Bayesian source apportionment of sedimentary organic carbon along a sluice-regulated river–estuary continuum: Coupling machine learning, spatial analytics, and multi-proxy geochemistry

Sedimentary organic matter (SOM) is a key integrator of aquatic biogeochemical processes under anthropogenic regulation, yet quantitative source apportionment in human-modified river–estuary systems remains limited. Here, a multi-proxy computational framework integrating grain-size energetics, Bayesian isotopic mixing (SIMMR), and landscape analytics was developed to quantify SOM dynamics along the gate-regulated Xinyanggang River–estuary continuum. Surface sediments ( n = 20) were characterised for textural parameters, elemental composition (TOC, TN, C:N mass ratio), stable carbon isotopes (δ 13 C), and watershed landscape metrics. A machine learning model (Random Forest) identified key environmental predictors explaining 82.69% of TOC variance, with total nitrogen, distance from the sea, and grain-size skewness emerging as primary drivers. A pronounced longitudinal energy gradient was observed from low-energy depositional conditions upstream to higher-energy regimes seaward, with discrete hydraulic discontinuities imposed by sluice regulation. Bayesian source apportionment using δ 13 C–C:N tracers revealed spatial partitioning of organic carbon origins: riverine plankton dominated upstream reaches (25–62%), anthropogenic inputs (soil + sewage: >30%) peaked in midstream areas, and marine-derived material (15.5–30.2%) increased markedly downstream of the gate. Landscape configuration metrics, particularly the aggregation index (which measures how strongly similar land-cover patches are spatially clumped), were significantly correlated with sedimentary carbon distribution patterns ( p < 0.05), indicating watershed structural controls on carbon delivery pathways. This integrative computational approach shows how routinely measured variables combined with advanced analytics can support data-driven assessment of gate operation impacts and watershed-scale carbon management in anthropogenically modified coastal river systems. • Bayesian mixing (SIMMR) resolves organic carbon sources with quantified uncertainty. • Random Forest highlights key drivers, explaining 82.7% of TOC variability. • Integrated multi-proxies reveal gate-driven texture–carbon interactions. • Spatial analytics link landscape patterns to sediment carbon accumulation. • Computational framework informs data-driven estuarine management strategies.