Metadata in Synthetics Calculation with FK in MTUQ

[SeismoFelix]

Hello,

I am facing an interesting situation when I am calculating synthetics in MTUQ with an FK database. Basically with these two lines I am doing all the job:

pre_synthetics = greens.map(process_sw)
synthetics = pre_synthetics.get_synthetics(mt, components=['Z', 'R', 'T'])

The synthetics calculation worked. However, when I tried to save the synthetics, I realized that the streams in synthetics list do not contain most of the metadata that is required for obtaining the station name, network, station location, etc. Interestingly, this only happens in the newest version of MTUQ. When I tried with an old version, the synthetics contains a comprehensive amount of metadata.

With old version:
print(pair[0].stats)

           network: IR
           station: TNSJ
          location: 
           channel: Z
         starttime: 2017-12-01T02:31:44.000000Z
           endtime: 2017-12-01T02:34:13.980000Z
     sampling_rate: 50.0
             delta: 0.02
              npts: 7500
             calib: 1.0
           _format: SAC
        depth_in_m: None
    elevation_in_m: None
                id: IR.TNSJ.
          latitude: 33.9633
         longitude: 56.6089
               sac: AttribDict({'delta': 0.02, 'depmin': -0.29059753, 'depmax': 0.22340882, 'b': -60.0, 'e': 555.98, 'o': 0.0, 'stla': 33.9633, 'stlo': 56.6089, 'stel': 1123.0, 'stdp': 0.0, 'evla': 30.734, 'evlo': 57.39, 'evdp': 6.0, 'dist': 365.57446, 'az': 348.60342, 'baz': 168.18732, 'gcarc': 3.2871227, 'depmen': 1.4267704e-06, 'cmpaz': 348.60446, 'cmpinc': 0.0, 'nzyear': 2017, 'nzjday': 335, 'nzhour': 2, 'nzmin': 32, 'nzsec': 44, 'nzmsec': 0, 'nvhdr': 6, 'npts': 30800, 'iftype': 1, 'idep': 7, 'iztype': 9, 'leven': 1, 'lpspol': 0, 'lovrok': 1, 'lcalda': 1, 'unused23': 0, 'kstnm': 'TNSJ', 'kevnm': '', 'kcmpnm': 'BHR', 'knetwk': 'IR', 'kinst': 'Tr40'})

With new version:

     network: 
     station: 
    location: 
     channel: Z
   starttime: 2017-12-01T02:33:31.129997Z
     endtime: 2017-12-01T02:36:01.109997Z

sampling_rate: 50.0
delta: 0.02
npts: 7500
calib: 1.0
_component: T
_format: SAC
processing: ["ObsPy 1.4.1: detrend(options={}::type='demean')", "ObsPy 1.4.1: detrend(options={}::type='linear')", "ObsPy 1.4.1: taper(max_length=None::max_percentage=0.05::side='both'::type='hann')", "ObsPy 1.4.1: filter(options={'zerophase': False, 'freqmin': 0.030303030303030304, 'freqmax': 0.06666666666666667}::type='bandpass')"]
sac: AttribDict({'delta': 0.1, 'b': 48.23, 'e': 457.73, 'o': 0.0, 'dist': 366.0, 'nvhdr': 6, 'npts': 4096, 'iftype': 1, 'iztype': 11, 'leven': 1, 'kevnm': ''})

---

In both cases, I used the same script and the same data, I only changed the MTUQ environment. On the other hand, this problem is not seen when I use Specfem3D GFs database. I also noticed that the starttime and endtime are also different despite I used the same origin time in both cases: '2017-12-01T02:32:44.000000Z'

If you can shed some light on what happened wit the .get_synthetics subroutine in the most recent version, I would appreciate it a lot. Without the metadata, it is difficult to figure out to what station corresponds a given synthetic trace.

Thanks for your help,

Felix

[SeismoFelix]

Probably this commit may be helpful for this discussion: 7668898 . Thanks @thurinj

[rmodrak]

Hi Felix, Thanks for the report. Let me check more later, but here at first glance is what I think is going on. Previously the metadata "stats" from the observed data were getting carried over to the synthetics. This works well many times, but doesn't generalize to all cases/ data formats. Now the metadata from the Green's functions themselves get carried over to the synthetics. The Green's function metadata is just what is supplied when you call the get_greens_tensors() method. At least this was the desired default behavior, let me check. You could also override the default behavior above and control directly what metadata gets passed to the synthetics via the "stats" argument to get_synthetics().

…

-Ryan

On Wed, Jan 15, 2025 at 4:18 PM Felix Rodriguez Cardozo < ***@***.***> wrote: Probably this commit may be helpful for this discussion: 7668898 <7668898> . Thanks @thurinj <https://github.com/thurinj> — Reply to this email directly, view it on GitHub <#291 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABCGSSSPORMGYNGWAFCGUW32K3UDJAVCNFSM6AAAAABVGSS3RGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKOJUGEZTAOJVGY> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[rmodrak]

Hi Felix, You mentioned a specific commit (7668898). Could you double check that you are using this version or a more recent one? I am finding that for such versions, the synthetic traces have the requested metadata

[rmodrak]

For reference, I used the current upstream version (f3d6cc7) and added these two lines to SerialGridSearch.DoubleCouple.py:

    synthetics_sw = greens_sw.get_synthetics(best_mt)
    print(synthetics_sw[0][0].stats)

The resulting standard output is

         network: YV
         station: BIGB
        location: 
         channel: Z
       starttime: 2009-04-07T20:11:15.360003Z
         endtime: 2009-04-07T20:14:05.340003Z
   sampling_rate: 50.0
           delta: 0.02
            npts: 8500
           calib: 1.0
      depth_in_m: None
  elevation_in_m: None
              id: YV.BIGB.
        latitude: 61.5919
       longitude: -149.8174

I also tried adding

    synthetics_sw = greens_sw.get_synthetics(best_mt, components=['Z','R','T'])
    print(synthetics_sw[0][0].stats)

and got the same result.

[rmodrak]

So at least get_synthetics() seems to give the requested metadata in examples/SerialGridSearch.DoubleCouple.py

If the behavior still continues in your case, could you post fuller example?