This repository contains the simulation code used in the following paper : Landreau, Mathieu, Boris Conan, Benjamin Luce, et Isabelle Calmet. 2025. « Observation and LES modeling of an offshore atmospheric undular bore generated during sea-breeze initiation near a peninsula ». https://hal.science/hal-04996636.

This is a fork of the "WRFlux" available at: https://github.com/matzegoebel/WRFlux which is a fork of the "Weather Research and Forecast model (WRF)" available at: https://github.com/wrf-model/WRF.

WRFstats allows to output time-averaged quantities, and 2nd and 3rd order products of quantities to be able to compute variances, and covariances in post-processing. Other small modifications have been made such as :

• A bug fix in tslist
• A feature to allow the time series maximum height to be different in each domain
• A feature ('fine_LU') that initialize soil properties by interpolating from parent domain (in case a nested domain starts later than its parent)
• A feature to chose the default soiltype that is used in land cells where the soil category is not defined.
• A bug fix (probably not the best way) wrf-model/WRF#1947

The diverse branches corresponds to the different implementation (stats for statistics), and run is the merge of all branches used for the paper (commit 7b8cf696)

During the model run, fluxes, tendencies, and budget variables are averaged over time. The online calculations can be controlled in the namelist file or the registry file Registry/registry.wrfstats. The calculations do not affect the model evolution.

The namelist variable output_stats is added. It can take the following values :

• 0 (default): no statistics is computed
• 1: return only 1st order statistics (mean)
• 2: return 1st and 2nd order statistics (mean, non-centered variances, non-centered covariances)
• 3 (or more): return 1st, 2nd, and 3rd order statistics

All variables are output to the auxiliary output stream auxhist24. The output interval can be set with the namelist variables auxhist24_interval_m and auxhist24_interval_s. The averaging period is equal to the output interval

The averaged variables can be found in Registry/registry.wrfstats, and the calculation is found in dyn_em/module_statistics.F. To add new variables, just add them in Registry/registry.wrfstats, and add their calculation in dyn_em/module_statistics.F

Variances and covariances can be computed in post-processing as

var(u) = u2_avg - u_avg*u_avg

covar(u, v) = uv_avg - u_avg*v_avg

This repository contains a complete, standalone version of WRF. Since it is a fork of WRF, the whole history of WRF's master branch is included as well. Thus, you can simply merge your changes with the changes that WRFstats introduces using git.

The calculation of time-average values is basic:

$\overline{x}_n = \sum_{i=0}^{i=n}(x_i) / n$

Then :

$\overline{x}_0 = 0$

$\overline{x}_{n+1} = (n*\overline{x}_n + x_{n+1}) / (n+1)$

It is computed in dyn_em/module_statistics.F. All the variables are averaged in the cell centers as in the timeseries feature. Velocities are interpolated before being averaged. This is mandatory to compute covariances.

The calculation of the water vapor density $\rho_d$ is based on Skamarock et al. (2019), Equation 2.16. The only averaged mixing ratio is the water vapor. The TKE_AVG variable is the time-averaged of either the TKE_PBL (from MYJ PBL model for example) or the TKE variable (subgrid TKE from TKE1.5 model). For some variables (e.g. pressure, temperature), the perturbation is averaged rather than the total quantity

The following WRF source code files have been modified:

• README_WRFlux.md (just changed the name)
• Registry/Registry.EM_COMMON (add TKE to the history)
• Registry/registry.em_shared_collection
• Registry/registry.wrfstats (new)
• dyn_em/Makefile (add dependency with module_statistics)
• dyn_em/depend.dyn_em (add dependency with module_statistics)
• dyn_em/module_initialize_real.F
• dyn_em/module_statistics.F (new)
• dyn_em/solve_em.F (stats)
• dyn_em/start_em.F (stats)
• main/depend.common (add dependency with module_statistics)
• share/interp_fcn.F (fine_lu)
• share/mediation_integrate.F
• share/module_check_a_mundo.F
• share/start_domain.F (fine_lu)
• share/wrf_timeseries.F (fix bug)
• wrftladj/depend.wrftladj
• wrftladj/solve_em_ad.F (stats)
• wrftladj/solve_em_tl.F (stats)

• WRFlux can increase the runtime by about 25 %. No test have been realized for WRFstats but the writing time might be highly impacted
• The tool does not work with adaptive timestepping

The online calculation have been compared to time series output

Feel free to report issues on Github. You are also invited to fix bugs and improve the code yourself. Your changes can be integrated with a pull request.

Thanks to Matthias Göbel who provided the code used as a starting point for WRFstats, and WRF developers and support for their amazing work.