Inconsistent Results on macOS M1 vs. Linux

[ronnychou]

Description:
I am encountering inconsistent results when running the LBLRTM simulation using the provided TAPE5 example in run_examples/run_example_built_in_atm_upwelling on macOS with M1 chip compared to running the same simulation in a Docker container on Linux/Ubuntu. The results from the Docker Linux environment appear to be correct, while the macOS M1 results seem to have an issue, maybe with the handling of surface temperature.

Steps to Reproduce:

1. Compile LBLRTM and LNFL using the linuxGNU option (e.g., make -f make_lblrtm linuxGNUdbl for LBLRTM and make -f make_lnfl linuxGNUsgl for LNFL). While the osxGNU option successfully compiles on macOS, the resulting LBLRTM output files cannot be read due to byte arrangement issues.
2. Run the LBLRTM simulation using the provided TAPE5 file on macOS M1.
3. Run the same simulation in a Docker container on Linux.
4. Compare the results.

Expected Behavior:

The results should be consistent across both platforms.

Actual Behavior:

The results differ significantly, with the macOS M1 output showing notable discrepancies in the window region (8-10 µm or 1000-1200 cm⁻¹). The brightness temperatures at weak absorption lines are much lower than expected, whereas they should theoretically be close to the surface temperature. This suggests a potential issue with the handling of surface temperature in the macOS M1.

Attachments:

Comparison plots

The LNFL TAPE5 file

$
   615.000  2030.000
11111111100000000000000000000000000000000000000    LNOUT EXBRD                             
%

The LBLRTM TAPE5 file used for the simulation.

$ Simple TAPE5
 HI=1 F4=1 CN=1 AE=0 EM=1 SC=0 FI=0 PL=0 TS=0 AM=1 MG=0 LA=0 OD=0 XS=0   00   00
 1000.0000 1200.0000       0.0                 0.0       0.0    0.0000     0.000    0            0.0
  294.2000    1.000  
    6    2    0    0    0        
   100.000     0.000   180.000
-1.
%

Environment:

macOS Version: macOS Sequola Version 15.3.2

Processor: Apple M1 Chip

Ubuntu Version: 24.04

LBLRTM Version: 12.17

LNFL Version: 3.2

Please let me know if any additional information or files are needed to investigate this issue further.

[ronnychou]

Additional Information:
I ran the Linux version of LBLRTM to simulate top-of-atmosphere (TOA) brightness temperatures for surface temperatures of 0 K, 100 K, 200 K, and 300 K. I plotted these results alongside the baseline output from the macOS M1 version of LBLRTM on the same graph. Additionally, I included a comparison of the differences between the Linux version (for 0 K and 100 K) and the macOS M1 version, which showed relatively small discrepancies. Notably, at 0 K, the differences were on the order of 10⁻¹³, appearing as a think flat orange line at the bottom of the graph. This further confirms that the macOS M1 version of LBLRTM has issues in handling surface temperature.

[pernak18]

@ronnychou weird MacOS is the problem. my initial instinct would have been to tell you not to trust the container since i last built that 5 years ago with an older version of the continuum and spectroscopy. you mention LNFL v3.2, but what line file are you using?

i've included other developers that might be able to do some testing hand help troubleshoot

[ronnychou]

@pernak18 @mjiacono @jrmascio Thank you all for looking into this issue. Just to clarify, I only used Docker as a convenient way to run Linux on my macOS system for comparison testing. All my actual builds were compiled from source code using the latest available versions:

• LBLRTM v12.17
• LNFL v3.2
• MT_CKD 4.3
• Line file v3.8.1

The discrepancies appear specifically when running natively on macOS M1 silicon, while the Linux builds (Ubuntu 24.04 in Docker on macOS and Windows) produce consistent, expected results. This confirms the issue is specifically with native execution on macOS M1 silicon, while containerized environments (regardless of host OS) all behave consistently. This suggests there may be platform-specific behavior in the macOS M1 implementation.

I'd be happy to provide any additional diagnostic information or run specific tests that might help identify the root cause. Please let me know what would be most helpful for your investigation.