Analysis of hyperspectral data to define multispectral sensor specifications for enhancing plastic greenhouse detection

Sentinel-2 and Landsat-9 multispectral operational missions enable global and continuous land-cover classification. These missions have been increasingly used to map plastic greenhouses; however, several misclassification issues were reported due to spectral similarities with other land covers. This research uses airborne hyperspectral data to demonstrate that plastic greenhouses can be reasonably mapped upon acquiring two reflectance values (ρ) in the Short-Wave InfraRed (SWIR) spectrum, at ρmax = 1644 nm and ρmin = 1729 nm, or alternatively at ρmax = 2130 nm and ρmin = 2315 nm. Consequently, we developed a Normalized Difference Greenhouse Index, NDGI = ρmax−ρminρmax+ρmin, for each alternative selection. These new indices rely on the spectral slope between characteristic wavelengths in the SWIR region rather than on the absolute depth of the absorption feature as a possible improvement to the unexplored Hydrocarbon Index (HI). The NDGI classified plastic greenhouses in a heterogeneous landscape with an accuracy of 93 % using high-resolution data. Still, calculating the NDGI with PRISMA ( PRecursore IperSpettrale della Missione Applicativa ) imagery revealed the method's transferability to coarser spatial resolutions despite a reduced separability inherently emerged. Furthermore, PRISMA data revealed that incorporating the bands for the NDGI would substantially reduce the number of false positives when using the Sentinel-2 based Open field and Protected Agriculture Classifier in the presence of critical highly reflective land covers. This study indicates that NDGI will enable efficient, fast, and operational mapping of plastic greenhouses globally through current hyperspectral missions (e.g., PRISMA by ASI, EnMAP by DLR, HISUI by JAXA, and Tanager by Planet) including the upcoming ones (e.g., NASA's Surface Biology and Geology mission – SBG and ESA's CHIME). Likewise, upgraded multispectral missions such as Landsat Next and Sentinel-2 Next Generation (NG) could integrate the identified spectral bands and allow for operational monitoring of plastic greenhouses.