We investigate how creep-driven convergence after cavern abandonment compresses trapped brine and alters geomechanical risk. Using the open-source finite-element simulator SafeInCave, we implement a two-way coupling between cavern volume and hydrostatic brine pressure and run fully coupled simulations for a field-scale cylindrical cavern whose roof lies between 600 m and 2200 m. Two abandonment protocols are considered: hard shut-in, in which we permanently seal the well, and soft shut-in, in which we vent brine whenever its pressure reaches 70% of the overburden stress. Each depth-protocol pair is simulated with and without pressure-solution creep (PSC).

After 300 yr a hard-shut-in cavern loses only 0.30% of its initial volume at 600 m and 0.76% at 2200 m, yet shallow caverns approach the micro-fracturing threshold as brine pressure climbs to 95% of lithostatic pressure. Soft shut-in preserves σ ≥ 5 MPa safety margin against microfracturing but allows greater closure: volume loss rises from 1.6% at 800 m to 2.4% at 1600 m before declining in deeper settings. Convergence initiates faster in deep caverns but decelerates below shallow-cavern rates as deviatoric stresses relax over time. Even in the worst case, 600 m depth under soft shut-in, surface subsidence reaches only 3.1 cm. PSC accelerates early convergence for both protocols; under hard shut-in its influence fades within decades, whereas the constant pressure offset under soft shut-in sustains PSC for centuries, adding approximately 0.8 cm of subsidence at 800 m but slightly reducing it below 1000 m.

Depth therefore governs post-closure behaviour. Caverns shallower than approximately 1 km experience rapid pressure build-up that pushes brine pressure to within 1 MPa of lithostatic stress, while deeper caverns become self-limiting and converge slowly. Above 1 km the shut-in protocol dominates risk, whereas below 1 km brine-pressure feedback controls the response. The key findings are:

Soft shut-in is essential for caverns with roofs shallower than approximately 1 km.

Hard shut-in suffices at greater depth, as overburden filtering limits surface subsidence to the centimetre range.

Future models should incorporate a stress threshold for PSC and stratigraphic layering to avoid over-predicting far-field deformation and rebound.

The coupled-physics workflow developed here thus offers regulators and operators a transparent baseline tool to forecast post-closure deformation, tailor abandonment strategy to depth, and direct monitoring resources where they matter most.