Slides for the presentation "Seismic data inversion with curvelet denoising preconditioning" given at the 15th International Congress of the Brazilian Geophysical Society & EXPOGEF 2017, Rio de Janeiro, Brazil.
Summary
Seismic inversion methods are highly sensitive to noise present in the data set. The need to enhance the signal-to-noise ratio (SNR) motivates many researchers do develop increasingly sophisticated denoising methods and combine them into other techniques. While some methodologies operate on a single scale, the curvelet transform established itself as multi-scale transform useful to decompose the seismic signals into multi-resolution elements. In this study, we evaluate the benefits of curvelet denoising as a preconditioning method for poststack seismic data in a 2D acoustic inversion process using a Bayesian framework. Our tests on a synthetic data set simulated in the Marmousi model, and a real data set from the Campos Basin offshore Brazil have shown that the curvelet thresholding method can be successfully applied for random noise elimination. Even the use of a hard global threshold might allow improvements in the deepest parts. Future work will have to show whether alternatives that ensure a more robust way of selecting the coefficients can take into account the wavelength change with depth.
Summary
Seismic inversion methods are highly sensitive to noise present in the data set. The need to enhance the signal-to-noise ratio (SNR) motivates many researchers do develop increasingly sophisticated denoising methods and combine them into other techniques. While some methodologies operate on a single scale, the curvelet transform established itself as multi-scale transform useful to decompose the seismic signals into multi-resolution elements. In this study, we evaluate the benefits of curvelet denoising as a preconditioning method for poststack seismic data in a 2D acoustic inversion process using a Bayesian framework. Our tests on a synthetic data set simulated in the Marmousi model, and a real data set from the Campos Basin offshore Brazil have shown that the curvelet thresholding method can be successfully applied for random noise elimination. Even the use of a hard global threshold might allow improvements in the deepest parts. Future work will have to show whether alternatives that ensure a more robust way of selecting the coefficients can take into account the wavelength change with depth.