Towards advanced mapping of plastic greenhouses from EMIT, EnMAP and PRISMA hyperspectral missions

Plastics are pivotal in extensive agricultural activities contributing towards the targets of the United Nations Sustainable Development Goal 2 (UN SDG 2). However, there are rising concerns about biodiversity changes and waste management challenges when plastics are used in agriculture that affect the targets proposed in the UN SDGs 12 and 15. Over the years, the general mapping of plastic greenhouses has been achieved using high spatial and multispectral resolution satellite missions. However, multispectral missions have limited information content and are prone to spectral shape ambiguities that limit the definitive identification of plastic greenhouses in natural environments with many heterogenous optically active targets. To this end, the current study proposes a verifiable workflow for a diagnostic spectral shape-based identification of plastic greenhouses utilizing open access hyperspectral imagery from ASI PRISMA, DLR EnMAP and NASA EMIT missions. A feasibility exercise was conducted in the Spanish province of Granada whereby the validation of observations including spectral characterisation of the greenhouses was achieved by proximal laboratory and airborne measurements. Polymer type of the fragments from the plastic greenhouses and harvested waste was revealed to be Low Density Polyethylene (LDPE). Identification algorithms for the LDPE plastic greenhouses were based on the diagnostic absorption features (∼1215, ∼1730, ∼2312 nm) found in the measured and continuum removed reflectance. Thematic maps and diagnostic optical features of the evaluated unique targets indicated the bottom-of-atmosphere reflectance Analysis Ready Data from the three satellite missions possessed consistent spectral shape similarities in the discrete images from 2021 to 2025. Matches in the generated maps suggested the algorithms were interoperable among the tested hyperspectral satellite imagery. The transferability potential of the proposed methods to other environmental scenarios or geographic regions (i.e., The Netherlands, Tunisia, Italy, Türkiye) was examined through a spectral-based inference approach. Insights were also presented on the added-value of having hyperspectral data as a way to mitigate the likely spectral ambiguities in algorithms based on the multispectral Sentinel-2 observations. The experimental findings also echo the benefits of exploring secondary applications and new variables from hyperspectral missions leveraging the vast information content that can be deciphered in the recorded big data.