When a person goes missing and a crime is suspected, police forces may scout for clandestine graves in certain areas in which the victim is suspected to have been buried. Such clandestine graves are often shallow and may affect the vegetation on top and around the grave for prolonged periods of time. Search efforts by visual means are only effective in detecting the effects of graves on vegetation when the effects are very pronounced. However, it is hypothesized that even when a visual observer cannot discern the effect from natually occurring variance, the spectral characteristics of the vegetation may still be sufficiently different to be able to be picked up through chemometric methods. This study aimed to explore if and how spectral imaging could be used to discriminate between the vegetation growing on shallow graves and on nearby reference plots. Data were acquired in the visible/near-infrared (Vis-NIR, 400–900 nm) and short-wave infrared (SWIR, 1000–1700 nm) with mobile spectral imaging set-ups and used to train and validate spectral classification models. 

For the duration of this project, we had access to the forensic human burial site of the Amsterdam University Medical Center. Regular visits were made to the burial site with mobile high-resolution spectral camera set-ups (featuring imec Snapscan VNIR and SWIR). During each visit, an older grave (~2-3 years) was fully imaged, as well as an adjacent reference plot where no body was buried. Both the grave and the reference plot were imaged in a mosaic-like manner, with some limited overlap between adjacent tiles. In January 2022, a new body was donated and buried in another part of the burial site. Also here, an adjacent reference plot was identified, but this time the plot was burrowed in a similar fashion as the grave. From February until May, monthly data acquisition runs were executed, during which both the older and the fresh grave, as well as the two reference plots, were fully imaged using Vis-NIR (400–900 nm) and SWIR (1000–1700 nm) spectral cameras. Small Spectralon white references (99% nominal reflectance) were kept in the field of view to be able to make proper corrections to the variable lighting.

All files in the dataset start with a date (YYYY-DD-MM) and wavelength range (VNIR and SWIR). File names then continue with a designation graf (grave) or referentie (reference plot) and either carry the subsequent designation oud (old) or not. Therefore, in file names graf_oud refers to the older grave, while referentie_oud refers to the reference plot adjacent to the older grave. Files without the oud designation were recorded on the newer grave or the reference plot adjacent to it. 

Spectral images and the corresponding metadata can be read with the spectral python library.