Rapid retrieval of soil surface aggregation as a joint attribute to soil spectral libraries

• RGB microscopy + SAM predicts soil surface aggregation rapidly. • Digital diameters correlate well with sieving (R 2 = 0.70, RMSE = 0.111 mm). • Spectral variation linked to surface aggregation and cementation agents. • RGB + SAM enhances physical realism of soil spectral libraries. • Scalable, low-cost proxy for surface aggregation in soil assessment. Utilizing Soil Spectral Libraries (SSLs) enables rapid, non-destructive, predictions of soil attributes by linking wet-chemistry data with the reflectance spectra of air-dried, <2 mm soil samples. However, the influence of surface aggregation on the light-scattering behavior of soil is generally overlooked primarily due to the complexity of its measurement, although it remains a key physical chromophore for databases. Yet, because surface aggregation can significantly affect spectral responses, particularly through its impact on scattering, it may represent a critically important parameter. Here, we present a novel, scalable and easy method combining USB-microscope RGB imaging and segmentation of the surface soil grains using the Segment Anything Model (SAM, ViT-H) a promotable, zero-shot vision transformer that infers surface aggregate size sensu strictu (<250 µm) distributions immediately prior to spectral measurement. Ninety-one air dried soils representing six USDA soil orders were sieved to <2 mm, imaged in five replicates under controlled illumination, and physically fractionated via quantitative laboratory-based sieving into six size classes (<0.1 mm to 2 mm) to compute gravimetric average aggregate diameter (AVG). RGB images acquired through a USB-microscope were segmented, filtered by contour properties, and particle diameters extracted via Feret diameter; aggregates <0.10 mm (derived from a measured µm px −1 ) were excluded. Mean, percentile, and error metrics were calculated per soil type. Correlation between digitally inferred RGB and gravimetrically laboratory derived aggregate diameters was strong (R 2 = 0.70, RMSE = 0.111, MAE = 0.091), with particularly robust performance in sandy and loamy soils, while clay-rich soils exhibited deeper subsurface heterogeneity. These results suggest that spectral standard deviation (SD) and variation in spectral replications of clayey soils may be higher than in sandy and silty soils which arises from the irregularly of the soil grain size, signaling to take precaution. Multi-angle imaging with increased replications aligns with standard spectroradiometer protocols and can be integrated into SSL curation pipelines. Both the physical and digital analyses of the soil’s grain size represent spectral correlation with the main cementation agents in the soil: Clay and SOM content. By embedding surface aggregation into SSLs, this method enhances the physical realism of proximal-sensing models, offering a cost-effective, time-efficient alternative to traditional physical sieving and advancing both laboratory and potential in-situ soil assessments.