Abstract
Conventional convolution-based seismic impedanceinversion usually requires prior wavelet information from welllog data or assumes a zero- or minimum-phase seismic wavelet.These requirements limit its applicability when well-log dataare unavailable or when the seismic wavelet exhibits mixedphase characteristics. To address these limitations, we proposea seismic-data-driven impedance inversion method based on preconditioned bidirectional deconvolution. Our proposed methoduses bidirectional deconvolution to estimate the seismic waveletdirectly from seismic data by decomposing it into minimumand maximum-phase components, which enables mixed-phasewavelet estimation without the need for zero- or minimumphase assumptions in previous approaches. The bidirectionaldeconvolution further incorporates a hybrid norm to promotesparse and spiky reflectivity, which is consistent with the blockynature of impedance. Using the estimated wavelet, we formulatean objective function that combines a data-fitting term linkingimpedance to seismic data with regularization on the impedance.We introduce depth-varying total variation (TV) regularizationto enhance impedance blockiness while adaptively weakening theconstraint at greater depths, where seismic amplitudes and datareliability decrease. We further use a 2-D Gaussian smoothingbased low-frequency constraint to recover missing low-frequencycomponents by matching the smoothed impedance to a constructed background model. During optimization, we adopt apreconditioning scheme to accelerate convergence. Under relatively homogeneous geological conditions, the impedance modelgenerally varies smoothly along stratigraphic layers. Therefore,we employ a local dip filter as the preconditioning operator toguide model updates along the local structural direction andthereby enhance the lateral continuity of the inverted impedance.Synthetic experiments on a dipping-layer model under noisefree, random-noise, and migration-swing conditions show that theproposed method recovers blocky impedance, preserves lateralcontinuity, and remains robust to noise and coherent artifacts.The Marmousi experiment further demonstrates its stabilityand applicability in complex geological settings with faults andintricate structures. Application to deepwater field data producesimpedance results consistent with well-log measurements andpreviously published inversion results, confirming the method’sreliability and practical applicability.
Paper Information:
L. Wang, Y. Shen, W. Peng and J. Chen, Seismic Impedance Inversion Using Preconditioned Bidirectional Deconvolution and Its Application to the Deepwater Field, in IEEE Transactions on Geoscience and Remote Sensing, vol. 64, pp. 5920118-5920118, 2026, Art no. 5920118, doi: 10.1109/TGRS.2026.3732373.

