Lab 3: Seismic Noise Attenuation

Seismic data are highly damaged with noise and unwanted energy came from different kind of sources. This noises can be classified into two main categories:1.      Random noise (incoherent noise)
2.      Coherent noise
However, it is impossible to eliminate all the noises, instead our goal ts to improve as much as possible the signal-to-noise ratio (SNR)

Noise

Ø  Random Noise
Disturbances in seismic data which lack phase coherency between adjacent traces are considered to be random noise. This energy is usually not related to the source that generates the seismic signals. Some examples of random noises in land records are near-surface scatterers, wind, rain, and instruments. Random noises can be attenuated easily in several different ways such as frequency filtering, deconvolution, wavelet denoising, filetring using Gabor representation, stacking and many other methods.

Ø  Coherent Noise
It is the vice versa of random noise, which is the energy comes from the source itself and considered undesirable energy that comes along with the primary signals. Such energy shows consistent phase from trace to trace. Examples of coherent noise in land records are multiple reflections or multiples, surface waves like ground roll and air waves, dynamite ghosts and others. Improper removal of coherent noise can affect nearly all processing techniques and complicates interpretation of geological structures

Ground Roll Noise

They are surface waves travelling along the ground surface and generally have low velocity, low velocity and high amplitudes. In time distance t-x curves, they are straight lines with low velocities and zero intercepts for an inline source. There might be several modes of ground rolls in the record because of their dispersive nature. They are attenuated using source and receiver arrays in the field and various processing methods such as frequency filtering and f-k filtering.

Ø  Spectrum Analysis and Filtering of Seismic Data
The filtering process is an important step in order to proceed further with other seismic data processing steps. Since our land seismic data set contains ground roll noise, we can simply apply frequency linear filters such as band-pass filters (BPF)’s to attenuate their effect. BPF’s in particular enhance the overall gain of each seismic shot gather, and increases the SNR ratio by attenuating low and high frequency noise records, including the ground roll noise

There exist different means for such useful analysis: the frequency content of 1D time signals, 2D like the frequency-space (f-x) or the frequency-wavenumber (f-k) spectra: each can be used to gain meaningful interpretation and, therefore, apply a suitable filtering technique. We can obtaion the f-x and f-k magnitude spectra in dB respectively by using MATLAB m-file fx.m and fk.m

Then, we can design a Finite Impulse Response (FIR) BPF digital filter and apply it to each seismic trace in the shot gather number 8 using the bpf_fir.m m-file as follow where we band the spectrum between 15 and 60 Hz and the remaining was left out.

1.    N=100
2.   Cut_off=[15,60];
3.   [Dbpf,Dbpf_f]=bpf_fir(Dshot,dt,N,cut_off);

           

Seismic data shot gather number 8 containing
ground roll noise before BPF filtering

Seismic data shot gather number 8 containing
ground roll noise after BPF filtering

Difference between before and after

a

b

c

             

Figure: The f-x magnitude spectra of the seismic data shot gather number 8 containing ground roll noise: (a) before and (b) after BPF filtering as well as for (c) the difference