International Journal of Plant & Soil Science

Carbon farming schemes increasingly pay land managers for increases in soil organic carbon, yet the credibility of these payments depends on monitoring, reporting and verification systems that can quantify small changes in a large and spatially variable carbon pool. Digital monitoring, reporting and verification promises to lower costs by combining satellite and airborne remote sensing, proximal soil sensing, artificial intelligence and process-based biogeochemical models, but the evidence that such combinations yield unbiased, verifiable estimates of soil organic carbon change remains uneven. This critical narrative review evaluates what each evidence stream can and cannot contribute to credible soil carbon accounting, how their integration propagates or reduces error, and which claims are established rather than preliminary. Literature was identified through structured searches of multidisciplinary scholarly indexes, backward and forward citation tracing and examination of institutional methodologies, and was appraised for design, validation strategy, uncertainty treatment and transferability. The synthesis indicates that direct measurement remains the only independent reference for soil organic carbon change, but that field-level detection is statistically weak unless sampling is dense, paired and replicated across many fields. Remote sensing is most reliable for documenting management practices and crop biomass rather than for retrieving topsoil carbon change, which is confounded by moisture, residue cover and roughness and is restricted to the surface layer. Proximal spectroscopy can reduce analytical costs substantially, provided that calibration transfer, bulk density and measurement error are handled explicitly. Machine learning improves spatial prediction but can overstate accuracy when validation ignores spatial dependence and extrapolation. Calibrated process-based models remain the principal engine for estimating change at scale, although independent time-series validation is scarce. Hybrid architectures that assimilate observations into mechanistic models are promising but have been tested mainly in a few temperate regions. Credibility therefore depends less on any single technology than on design-based verification, counterfactual baselines, transparent uncertainty propagation and explicit validity domains. Priorities include randomised multi-field measure-and-remeasure networks, open benchmark datasets, standardised uncertainty reporting and evaluation in smallholder and tropical systems.