Four basic pre-processing tools are used when constructing norm synthetic seismic data from raw simulated data. These four steps are:
In a partial Madagascar Sconstruct example as what shown in the following codeblock, sfbandpass, sfwindow are used for re-sampling, where sfbandpass are applied before down-sampling to avoid aliasing.
Then, sfpow is used for amplitude balance along the time axis, where attenuation caused by geometry spreading would be alleviated.
Afterwards, a time window is used to cut the data into the same size, which named as the height in common AI applications. Here we have the height 2048 points.
Finally, we normalize data by traces via sfscale to limit the values of all seismic data within negative to positive one.
# parameters
dt = 0.0005 # unit: s
fa = 0.1 # time point of first arrival on one trace
nt = 2048 # length of time axis
path = os.path.join(cwd, pown_dir) # local path to save files
# Madagascar function
Flow(out_name, in_name, '''
put o1=500 d1=20 label1=Depth unit1=m
o2=0 d2=%g label2=Time unit2=s |
transp plane=12 |
bandpass fhi=500 |
window j1=4 |
pow pow1=2 |
window f1=%g n1=%g |
scale axis=1
datapath=%s
''' % (dt, int(fa/dt/4), nt, path)
)
A complete workflow of constructing norm data from raw synthetic seismic data can be found at scripts.