Advancing soil texture and organic carbon spatial variability assessment: Integrating proximal γ-ray spectroscopy and electromagnetic induction via data fusion for site-independent analysis

In this study we tested the combination of two proximal sensors –an electromagnetic conductivity meter (EMI) and a gamma(γ)-ray detector– to estimate the variability of some soil properties. The hypothesis was that the employment of data fusion techniques would provide a more comprehensive description of soil texture and soil organic carbon (SOC) content compared to using single sensors alone. This approach aimed to capture a spectrum of characteristics across various mineral and organic alluvial soils. Field surveys covering more than 60 ha were conducted in 2019 on five agricultural experimental sites, whose texture ranged from silty-clay to silty- and sandy-loam and SOC from 0.5 % to 21.9 %. A total of 354 undisturbed soil samples (0–30 cm) was collected and soil properties such as granulometry, bulk density and SOC concentration and stock determined. The fusion of proximal sensing data of apparent electrical conductivity (EC_a) and γ-ray radionuclides (total counts –TC–, ²³⁸U, ²³²Th, ⁴⁰K) combined with multivariate analysis was applied to describe soil spatial variability. Partial leas square regression (PLSR) and artificial neural network (ANN) models were trained and tested to predict the variability in soil properties within and between sites. Results showed that data fusion captured the soil spatial variability within and between fields, with a predictive ability (test set) to explain up to 88 % of clay and 74 % of SOC stock variability when all mineral soils were embedded in the ANN model. The combination of TC and EC_a was particularly effective in explaining texture and SOC heterogeneity within and across fields. In contrast, different responses were observed for single radionuclides, either ²³⁸U, ²³²Th, ⁴⁰K, or their ratios, within each field, that likely identified site-specific radioisotope enrichment and/or depletion processes. In conclusion, data fusion of EMI and γ-ray detectors accurately predicted soil texture and SOC across soils from alluvial origin.