Bayesian modelling of salinity profiles in the Volturno River mouth under depth dependence

Saltwater intrusion is a recurrent environmental hazard for Mediterranean coastal systems, and identifying which physico-chemical variables track salinity is central to monitoring design. Estimating such associations from depth-resolved profiles requires two features of the data to be addressed: the uncertainty of each depth-bin summary, arising from pooling readings taken at several stations along a steep longitudinal gradient, and the residual vertical dependence left after conditioning on the covariates. This paper proposes a Bayesian errors-in-variables-like model with a Gaussian process component for residual depth structure, applied to two synoptic campaigns in the Volturno River mouth (Italy) under contrasting winter and summer hydrological conditions. The results show that temperature is consistently associated with salinity, but the sign of the association differs between the two seasonal conditions during sampling: it is positive in winter and negative in summer within the reach occupied by the salt wedge. Dissolved oxygen is important in winter only. The residual depth amplitude is small in both campaigns, while the complementary full river reach analysis for the summer campaign yields a substantially larger residual depth component, showing that the fitted relationship is sensitive to the sampled river reach. Under near-perfect collinearity the ordering of the covariates is better identified than any individual coefficient, and is reported through posterior rank probabilities. A simulation study shows that the specification maintains nominal credible-interval coverage where a naive regression collapses to a few per cent and reverses a tracer coefficient’s sign in up to 45 per cent of replicates. Cross-validation on held-out depth bins supports predictive adequacy within each campaign, while over comparable river reaches the estimated relationships are specific to the contrasting hydrological regimes represented by the two campaigns and do not transfer between them.