GDAS1 stability (vmixing) output differs ~0.5% between READY web and local run with identical parameters

[yestay]

I'm computing boundary-layer stability parameters (mixing height, Kz, u*, Kh) from GDAS1 archive data for a research project, using the vmixing program locally (HYSPLIT v5.4.2, public Linux binary, gfortran).

I compared my local output against the READY web "GDAS1 Stability" product for the same location (49.79 N, 73.17 E), same date (2019-01-01), and the same GDAS1 archive file (gdas1.jan19.w1). Parameters are identical (KBLS=1, KBLT=2, KMIXD=0, KMIX0=150, KZMIX=0).

The values agree very closely but not exactly. For example, at 2019-01-01 00 UTC:

Zi (m) 10xKz U* (m/s) Kh (m²/s)
READY web 150 30.65 0.1303 876.1
Local 150 30.84 0.1307 881.5

The differences are consistently ~0.3–0.6%. I've verified they are not caused by parameters (I scanned KBLS/KBLT/KMIXD/KMIX0/KZMIX), flags, coordinates, or the met file.

Questions:

Is this small difference expected, e.g. due to the READY server using a different compiler (ifort) or a different HYSPLIT build than the public Linux binary?
Is the READY web vmixing configured identically to the standalone vmixing program, or are there server-side differences?
For a large dataset (multiple sites, 2019–2024, 3-hourly) that cannot be requested manually through the web interface, is it acceptable/recommended to use the local vmixing output as equivalent to the READY web product?

Thank you.For all developers and scientist

[sonny.zinn]

READY runs the vmixing program for stability calculations. While minor discrepancies are expected due to single-precision floating-point arithmetic, a difference of 0.3–0.6% is unexpectedly large for the same input.

What is your top of domain height in meters? READY uses 10,000 meters. If the domain height is changed to 15,000 meters, numbers will change because of changes in vertical interpolation.

Zi (m) 10xKz U* (m/s) Kh (m²/s)
150 30.65 0.1303 876.1 (10,000 m)
150 30.84 0.1307 881.5 (15,000 m)