Differences in WRF simulations when initializing with ERA5 (pressure levels) specific humidity vs relative humidity

[wrf_marco]

Hello everyone,

According to the WRF User Guide and various forum discussions, WRF can be initialized using either relative humidity (RH) or specific humidity (q) as atmospheric moisture input. The documentation generally states that both approaches are equivalent, since real.exe converts the input humidity fields into QVAPOR in wrfinput and wrfbdy.

However, during some recent tests I noticed that the two approaches lead to noticeable differences in the resulting simulations, even though the workflow technically works in both cases.

Experiment setup​

• Model: WRF
• Domain: MED-CORDEX domain
• Resolution: 12 km
• Physics / namelist: standard MED-CORDEX configuration (already well tested)
• Simulation length: 1 month (January 2020)
• Output frequency: hourly
• Driving data: ERA5

Two simulations were performed with identical configuration, changing only the humidity variable used in the WPS preprocessing.

Case 1 — Relative Humidity​

• ERA5 RH (var157) used in WPS
• Default ECMWF Vtable (with RH)
• met_em files contain RH
• real.exe computes QVAPOR from RH

Case 2 — Specific Humidity​

• ERA5 specific humidity (var133) used
• Vtable manually modified to replace RH with
  133 | 100 | * | | SPECHUMD | kg kg-1 | Specific humidity
• met_em files contain SPECHUMD
• real.exe computes QVAPOR directly from q

In both cases the preprocessing chain runs correctly and produces valid wrfinput and wrfbdy files.

Result​

Despite this, the resulting simulations show systematic differences:

• Monthly mean temperature differences generally < 0.5 K, but locally exceeding 1 K.
• Accumulated precipitation differs noticeably across several regions.
• Instantaneous differences in temperature and precipitation appear throughout the simulation.

I will attach:

• maps of temperature differences: case2(spechumd) - case1 (RH)
• maps of accumulated precipitation differences: case2(spechumd) - case1 (RH)
• namelist.input
• Vtable.ECMWF with SPECHUMD (case2) and Vtable with RH (default, case1)

Let me know if you want to look at other variables in wrfout.

Question​

Since both workflows are described as equivalent in the documentation, I would like to ask:

1. Is it expected that using RH vs specific humidity as input leads to noticeable differences in WRF simulations?
2. Is there a recommended best practice when driving WRF with ERA5 (prefer RH or specific humidity)?
3. Could these differences arise from how interpolation and humidity conversion are performed in WPS/real.exe?

My initial expectation was that using ERA5 specific humidity should be theoretically preferable, since WRF internally uses QVAPOR. Using RH requires an additional conversion step (RH → q) within real.exe, which depends on temperature and pressure and may follow slightly different formulations than those used in ERA5. Therefore the workflow ERA5 → RH → recomputed q in WRF could introduce additional inconsistencies compared to directly using ERA5 specific humidity.

Finally, I suspect that even relatively small differences in humidity initialization could lead to systematic differences in long climate simulations (multi-year runs), especially through nonlinear processes such as cloud formation and precipitation.

I would be very interested to know whether others in the community have investigated this or observed similar differences.

Thank you very much for your support!

Best,
Marco