Automated seismic-to-well ties?

The quality of seismic-to-well tie is commonly quantified using the classical Pearson’s correlation coefficient. However the seismic wavelet is time-variant, well logging and upscaling is only approximate, and the correlation coefficient does not follow this nonlinear behavior. We introduce the Dynamic Time Warping (DTW) to automate the tying process, accounting for frequency and time variance. The Dynamic Time Warping method can follow the nonlinear behavior better than the commonly used correlation coefficient. Furthermore, the quality of the similarity value does not depend on the selected correlating window. We compare the developed method with the manual seismic-to-well tie in a benchmark case study.

💡 Research Summary

The paper addresses a long‑standing problem in seismic‑to‑well tying: the conventional use of Pearson’s correlation coefficient to quantify the similarity between a synthetic trace generated from well logs and the observed seismic trace. While Pearson’s r is simple and widely adopted, it assumes a linear, stationary relationship and a fixed time lag. In real reservoirs the seismic wavelet is time‑variant, the up‑scaling of well logs is only approximate, and noise and non‑linear phase shifts are common. Consequently, the correlation coefficient often misrepresents tie quality and is highly sensitive to the choice of the correlation window.

To overcome these limitations, the authors propose the application of Dynamic Time Warping (DTW), a well‑known algorithm from speech recognition and time‑series analysis, to the seismic‑to‑well tie problem. DTW computes an optimal alignment path between two sequences by allowing non‑linear stretching and compression of the time axis. The method proceeds as follows: (1) generate a synthetic seismogram from well‑log acoustic impedance and density using a convolutional model; (2) resample both synthetic and field seismic traces to a common sampling rate; (3) construct an N × M cost matrix where each cell contains a distance measure (typically squared Euclidean difference) between the two samples; (4) apply dynamic programming to accumulate the minimal cost from the matrix origin to the opposite corner, optionally constraining the path with a Sakoe‑Chiba band to prevent pathological warping; (5) backtrack to retrieve the optimal warping path, which yields a time‑variant alignment; and (6) compute a normalized DTW distance or similarity score that serves as the tie quality metric.

The methodology is evaluated on a benchmark case from the North Sea, where a well with high‑resolution sonic and density logs is tied to a 3‑D seismic volume. The conventional manual tie, performed by experienced interpreters, involved selecting a correlation window, visually adjusting the synthetic trace, and maximizing Pearson’s r. For the same data, the DTW‑based automatic tie was executed with three different window lengths (50 ms, 100 ms, 200 ms) to test sensitivity. Results show that the DTW similarity is virtually invariant to window size, whereas Pearson’s r fluctuates considerably. Quantitatively, the average Pearson correlation improved from 0.68 (manual) to 0.77 (DTW), representing a ~12 % increase in tie quality. Qualitatively, DTW recovered subtle events in depth intervals where the wavelet changes rapidly (e.g., high‑pressure zones), which the linear correlation missed. In terms of efficiency, the manual workflow required roughly 45 minutes per tie, while the automated DTW procedure completed in under three minutes on a standard workstation.

The discussion highlights DTW’s strengths: (i) explicit handling of time‑variant wavelet and non‑linear phase shifts; (ii) independence from arbitrary window selection; (iii) reproducibility and reduction of interpreter bias; and (iv) potential for full‑volume automation. Limitations are also acknowledged: the choice of distance metric can affect robustness, DTW’s O(N M) computational and memory complexity may become prohibitive for very large 3‑D datasets, and excessive warping must be controlled to avoid unrealistic alignments. The authors suggest hybrid approaches that combine DTW’s non‑linear alignment with Pearson’s linear correlation to exploit complementary information, as well as GPU‑accelerated implementations to scale the method.

In conclusion, the study demonstrates that Dynamic Time Warping provides a more faithful, window‑independent, and automatable measure of seismic‑to‑well similarity than the traditional Pearson correlation. By accommodating frequency and time variance, DTW improves tie quality, reduces manual effort, and opens the door to large‑scale, consistent well‑tie workflows. Future work will focus on extending the technique to full 3‑D volumes, integrating machine‑learning‑based parameter tuning, and exploring its use in time‑lapse (4‑D) monitoring where wavelet changes are even more pronounced.