seismic data conditioning

Seismic spectral balancing is a broadband enhancement technique that flattens the amplitude spectrum of seismic traces to compensate for source band-limitation, earth filtering, and frequency-dependent attenuation (Q-loss). The method works by dividing the data into overlapping time windows, computing each window’s smoothed amplitude spectrum, and constructing a frequency-dependent gain operator that boosts weaker high frequencies and moderates strong low frequencies. Applying this operator in the frequency domain and reconstructing the trace produces a more uniform (“white”) spectrum that improves vertical resolution and reflector continuity. 

Time-variant spectral balancing applies progressively stronger high-frequency boosting with depth to counter increasing attenuation. While effective and stable compared to physical Q-compensation, spectral balancing can amplify noise and requires parameter control (window length, smoothing, whitening power, stabilizer) to avoid over-whitening or amplitude distortion.

Amplitude gain in seismic processing is a family of techniques that compensate for time-dependent energy decay caused by geometric spreading, intrinsic attenuation, absorption, and transmission losses as waves travel deeper into the subsurface. Since deeper reflections naturally have lower amplitudes, gain functions such as AGC (Automatic Gain Control), true-amplitude spherical divergence correction, exponential or power-law time-variant gains, and structure-consistent scaling are applied to equalize amplitudes across the trace. These methods typically compute a windowed or global envelope, derive a gain factor that boosts weaker parts of the signal, and apply it in the time domain to restore relative visibility of deep events. Proper amplitude gain improves reflector continuity, enhances interpretability, and stabilizes subsequent processes like filtering and migration. However, excessive or poorly designed gain can distort true amplitude relationships, mask AVO effects, exaggerate noise, or create artificial continuity; therefore gain type, window length, and stabilization parameters must be chosen carefully to preserve geologic amplitude fidelity.

FX deconvolution is a frequency–space domain technique for suppressing random noise and enhancing coherent seismic events by exploiting the fact that true reflections vary smoothly across traces, while random noise is uncorrelated trace-to-trace. The method transforms each time window of the seismic section into the FX domain via FFT along the time axis; at each frequency slice, the complex amplitudes across traces are modeled as a predictable, low-order autoregressive (AR) process. The AR coefficients are estimated using least-squares or stabilized prediction-error filters, and the predicted coherent signal is reconstructed while the unpredictable energy random noise is attenuated. After inverse FFT to the time domain, the output section exhibits improved continuity and reduced random noise. FX deconvolution works best when events are linear and coherent, but requires proper AR model order, window size, and stabilizers; if poorly tuned, it can leave residual noise, distort dipping events, or introduce frequency-dependent artifacts, especially in low-SNR or aliased data.

A dip-guided median filter is a structure-oriented noise-reduction technique that combines the robustness of median filtering with the directional preservation provided by dip estimation. First, a local dip field is computed using methods such as plane-wave destruction or gradient-based slope analysis to determine the orientation of seismic reflectors at each sample. Using this dip field, the algorithm extracts a window of samples aligned along the reflector dip rather than in a vertical or fixed rectangular neighborhood. The median of the samples within this steerable, dip-aligned window is then assigned to the central sample, which effectively suppresses spike noise, burst noise, and small incoherent events while maintaining reflector continuity and protecting edges, terminations, and faults. Because the median operator is non-linear and resistant to outliers, dip-guided median filtering performs better than amplitude-based smoothing when random spikes or high-amplitude noise are present. However, its success depends on accurate dip estimation; poor dip fields or strong conflicting dips may cause reflector distortion or over-smoothing.