Skip to content

P3 modifications from Fortran P3

VVMex uses the Kokkos-based P3 microphysics scheme from E3SM EAMxx as the implementation framework, but modifies the process coupling to follow the original Fortran P3 behavior needed by the Vector Vorticity Model (VVM). The main change is that VVMex handles vapor-cloud-rain-ice phase adjustment inside the P3 microphysics step instead of relying on EAMxx macrophysics/SHOC to provide those tendencies externally.

In EAMxx, several processes are split across P3, SHOC, and other safeguards. In VVMex, these processes are coupled inside P3 so the microphysics can update vapor, hydrometeor mass, number, and thermodynamic state consistently during the model step.

Process comparison

Feature / process VVMex modification following Fortran P3 EAMxx P3 behavior
Cloud fraction All-or-nothing treatment. Cloud fraction is decided by SHOC.
Air density Full non-hydrostatic calculation, \(\rho = P / (R T)\). Hydrostatic approximation, \(\rho = -\frac{1}{g}\frac{dP}{dz}\).
Saturation vapor pressure (qvs, qis) Uses Polysvp1, combining Flatau et al. (1992) style polynomial saturation vapor pressure with Goff-Gratch behavior. Uses Murphy-Koop or Polysvp1; the Polysvp1 path is Flatau-only.
Droplet activation Performed after ice_nucleation and before cloud_water_autoconversion, together with the initial-step saturation adjustment. Coupled to SHOC or prescribed external inputs.
Rain droplet size distribution Marshall-Palmer distribution with \(\mu_r = 0\). Generalized Gamma distribution with \(\mu_r = 1\).
In-cloud limits No artificial in-cloud or precipitation thresholds are applied. Applies explicit incloud_limit = 5.1e-3 and precip_limit = 1.0e-2 thresholds.
Ice lookup table Uses v6.4. The table resolution differs, so the lookup_table_1a_dum1_c coefficient is changed accordingly. Uses v4.1.1.
Condensation, evaporation, deposition, sublimation Partitions vapor phase change among \(q_c\), \(q_r\), and \(q_i\) using relaxation rates and saturation state. SHOC handles condensation and evaporation of \(q_c\). Deposition, sublimation, and evaporation of \(q_r\) are handled separately in ice_deposition_sublimation() and evaporate_rain().
Bergeron process Included within the coupled mixed-phase vapor-adjustment step. Treated as an independent process tendency.
Mass conservation Applies coupled vapor-hydrometeor mass and number updates. Processes are updated separately.
Supersaturation limiter No extra EAMxx-style supersaturation limiter; saturation behavior is controlled through the coupled vapor-adjustment and conservation update. Additional safeguards prevent overshooting saturation.

Restored and modified kernels

The Fortran-P3-style logic is integrated into the Kokkos pack-oriented kernels under src/physics/p3/impl/.

p3_calc_cond_evap_dep_sub_impl.hpp

This file restores the semi-analytic integration for thermodynamic phase changes and combines competing vapor adjustment pathways according to their calculated timescales. It handles:

  • Cloud condensation/evaporation: explicit tendencies for \(q_v \rightarrow q_c\) (qv2qc_conden_tend) and \(q_c \rightarrow q_v\) (qc2qv_evap_tend).
  • Rain condensation/evaporation: vapor exchange involving \(q_r\).
  • Ice deposition/sublimation: restored processes for \(q_v \rightarrow q_i\) and \(q_i \rightarrow q_v\).
  • Mixed-phase competition: coupled vapor adjustment among \(q_c\), \(q_r\), and \(q_i\) based on relaxation rates and saturation state.

p3_droplet_activation_impl.hpp

This file restores droplet activation inside P3. It computes new droplets from environmental supersaturation and initializes both cloud droplet number tendency (nc_nuclet_tend) and associated mass tendency (qv2qc_nucleat_tend). In the VVMex sequence, activation occurs after ice nucleation and before cloud-water autoconversion.

p3_limit_cond_evap_dep_sub_impl.hpp

This file limits the coupled condensation, evaporation, deposition, and sublimation tendencies so the update does not over-deplete available water or violate the coupled mass/number update. In VVMex this is part of the Fortran-P3-style coupled vapor adjustment, not the same as the additional EAMxx supersaturation safeguard listed in the comparison table.

Main-loop integration

These processes are integrated into the p3_main execution flow. In p3_main_impl_part2.hpp, the sequence includes:

  1. Phase-change tendency calculation: calc_cond_evap_dep_sub computes coupled condensation, evaporation, deposition, and sublimation tendencies.
  2. Droplet activation: droplet_activation computes supersaturation-driven new droplet mass and number.
  3. Coupled limiting and conservation: limit_cond_evap_dep_sub scales the coupled tendencies before mapping back to cell averages.
  4. Prognostic updates: tendencies such as qv2qc_conden_tend, qc2qv_evap_tend, qv2qi_depos_tend, and related number tendencies are injected into the liquid and ice update paths to advance \(q_v\), \(q_c\), \(q_r\), \(q_i\), temperature, and potential temperature consistently.

Precision-specific lookup tables

p3_init caches the four non-ice tables (mu_r, vn, vm, revap) as raw binary dumps of the working scalar type, so the filename carries the element size: *_v2.dat8 for a double-precision build and *_v2.dat4 for a single-precision one. rundata/p3/ therefore has to hold both sets, and both are required inputs, not optional caches.

Nothing checks that the file was actually read. read_computed_tables streams the bytes straight into a zero-initialized Kokkos view, so a missing or truncated file leaves every table at zero and the run continues. With vn_table_vals zeroed the rain fallspeed is zero: \(q_c\) and \(q_r\) still form and look normal, but rain never reaches the ground and surface precipitation is effectively zero for the whole run. This is silent — there is no warning and no abort.

Regenerate a missing set by running any case once with the matching build and

{ "physics": { "p3": { "make_lookup_table": true } } }

This computes the tables in memory and writes them to rundata/p3/. Use a single rank because every rank writes the same paths. Set the option back to false afterwards.

The tables are computed by integrating the drop size distribution over 10,000 bins in the working precision, so a .dat4 set generated this way sits up to 2.3e-5 (relative) from the corresponding .dat8 values. That is well inside single precision's own error budget for the model, but it does mean the two sets are not simply rounded copies of each other.

Finding the VVMex changes

Start with p3_main_impl_part2.hpp to see the process order. Function declarations and dispatch live in p3_functions.hpp; the three implementation files named above contain the coupled tendency calculations, droplet activation, and limiting. Search for the tendency symbols such as qv2qc_conden_tend and qv2qi_depos_tend when tracing their updates.