This dataset contains digitised Pfeiffer chromatogram images obtained from soil samples of the Soil Fertility Laboratory Quality Analysis Program (PAQLF), coordinated by Embrapa Solos (Brazil) and organised as an interlaboratory proficiency test. The dataset was created to support the development and validation of end-to-end deep learning frameworks for the simultaneous quantitative prediction of 15 soil physicochemical properties from visual chromatographic patterns. Pfeiffer chromatography encodes the soil biogeochemical signature in radial visual patterns generated by organomineral interactions with a capillary membrane. Each chromatogram is digitised under two conditions: (i) white background (analytical image, used for modelling) and (ii) high-contrast green background (auxiliary image, used exclusively for automated geometric segmentation). A binary segmentation mask is derived from the auxiliary image and applied to the analytical image to produce a standardised region-of-interest representation. The dataset supports multitarget regression, comparative evaluation of deep learning architectures, and methodological studies on the use of sequential chromatographic diffusions as a temporal enrichment strategy under limited data conditions.