Convection-permitting modeling

Convection-permitting modeling means running a regional model at grid spacings of roughly 1-4 km and switching off the cumulus parameterisation, so that deep convection develops through the resolved dynamics.

The word permitting is doing careful work and is worth taking literally. Deep convective updrafts are on the order of a kilometre across, so a 3 km grid represents them with a handful of points at best. The model permits convection to occur without resolving its internal structure. The alternative term “convection-resolving” is used loosely but is not accurate at these resolutions, and the distinction matters when interpreting output — the cells a model produces at 3 km are systematically wider and fewer than real ones.

What it fixes

The improvements are consistent enough across studies to be considered established:

• The diurnal cycle of rainfall. Parameterised convection tends to peak near local noon; observations peak in late afternoon or evening. Convection-permitting runs correct this, largely because initiation now depends on the resolved flow rather than on a scheme’s trigger function and its handling of CIN.
• Rainfall intensity distribution. Parameterised runs rain too often and too weakly. Permitting convection produces a far more realistic distribution of intensities and, importantly, of dry hours.
• Organisation. Mesoscale convective systems, squall lines, and cold pools appear as coherent structures. Cold pools matter more than they might seem: they trigger new convection at their edges, and that mechanism is entirely absent from most parameterisations.
• Terrain and coastal effects. Sea breezes, valley circulations, and orographic triggering are represented far better, which is most of why the approach matters for complex terrain.

What it does not fix

The parameterisation problem is relocated, not removed. Turbulence, radiation, microphysics, and land-surface exchange all still require schemes, and two of them become more awkward at this resolution rather than less.

The grey zone is the substantive difficulty. At 1-4 km the model sits between the scales where a boundary-layer scheme is valid (turbulence entirely sub-grid) and where LES is valid (turbulence largely resolved). Neither assumption holds, and standard PBL schemes are applied outside their design range. This is a known, unresolved issue rather than a detail.

Microphysics also carries more weight. With convection resolved, precipitation comes entirely from the microphysics scheme, so its assumptions about drop size distributions and fall speeds now set rainfall intensity directly. Sensitivity to microphysics choice increases substantially.

And the cost is severe: a factor of roughly 100 or more against a 12 km run, once the shorter time step is included. That constrains convection-permitting climate simulation to short slices, selected episodes, or small domains — which makes episode selection a scientific decision, not a technical one.

See also: dynamical downscaling for the general approach, regional climate modeling for the context, convective rain for what it improves, and WRF for the model I have used for it.