Loading model data…
One hypothetical factory is placed in a mountain crop. Surrounding mountains and local buildings use the retained original scene with seed42 exaggerated relief. All original source emissions and initial background are excluded. Reused frozen GFS forcing starts4 November2026 at18:00UTC;36h weather includes12h adjustment, followed by24h continuous source transport. Counterfactual terrain scenario, not a geographic forecast.
Independent WRF-compatible JAX weather, approved Noah-MP configuration, and CPU FLEXPART transport. The qualified r04 model is retained. Earlier2mm CPU-consistency evidence is not an observation-error bound for this altered terrain.
This view shows one outlet-origin source increment, not total ambient SO₂. Initial-background particle propagation failed local concentration checks and is excluded. The near-source field is a 3D particle-residence reconstruction with parameterized thermal rise and entrainment, not resolved stack/building CFD. Wind varies with height and receives no SO₂ feedback. Fine-bin point values have Monte Carlo uncertainty. The changed terrain enters native weather initialization and3D transport. Visible mountains follow native model-ground, bilinearly subdivided without added terrain detail; original building geometry is redraped onto this surface. The level plant pad is display geometry only. Terrain altitude is displayed at6×. Outlet heights and concentration height above ground retain their3× scale; their bases follow the same displayed terrain. Native heights and numerical queries are unchanged. The1km weather and native stochastic turbulence do not resolve individual building wakes or small-eddy LES. Parent/child integration steps are3/1 seconds with epssm=0.5 acoustic time off-centering. This explicit numerical stability change passed native10min and GPU6h before same-carry continuation to36h. The terrain geometry is sampled from the prior3km HGT, so1km grid spacing does not add measured terrain detail. Land-cover/static fields and the local boundaries are resampled on the new grids from the same frozen scenario.
Equations, numerical results & limits ↗