Field-level spectral support auditing of legacy land-use inventories using PRISMA hyperspectral data

Regional land-use inventories are frequently used as reference information in agricultural remote sensing, although their labels may be temporally displaced relative to the satellite observations being analysed. This is problematic when legacy polygon labels are interpreted as contemporaneous pixel-level ground truth. We propose a field-level spectral-support audit for the 2016 SIT land-use inventory of Puglia, Southern Italy, using PRISMA Level-2D hyperspectral acquisitions from 2019–2025. The inventory labels are treated as weak operational priors, and the audit quantifies how strongly recent hyperspectral observations support each assigned original SIT class. For supported SIT LIVELLO_4 classes, robust spectral prototypes were estimated from a scene-balanced PRISMA–SIT export. Pixel spectra were compared with the prototype of their assigned class, transformed into spectral-support scores, and aggregated at field–scene level using robust statistics. The output is an inspection-priority layer that separates spectrally typical, intermediate-support, and spectrally atypical field–scene units, while retaining unsupported or too-small units with explicit flags. A targeted photo-interpretation analysis of 158 linked and eligible field–scene units retained from an original 170-case diagnostic sample showed that spectral atypicality should not be interpreted as semantic error: many spectrally atypical units were visually consistent with the SIT label, whereas some spectrally typical units were more consistent with alternative interpretations. In this diagnostic sample, semantic alternatives appeared more class-dependent than stratum-dependent. The proposed framework therefore does not provide a calibrated change-detection or accuracy-assessment product. Instead, it offers a traceable way for regional authorities to transform PRISMA observations into a ranked field-level queue for targeted inventory inspection and update.