Enhancing the Accuracy of Dynamic Time Warping by Integrating Stratigraphic Constraints Into the Automated Correlation of Sedimentary Sequences

Abstract Stratigraphic correlation of sedimentary archives is a crucial task in the field of paleoclimatology. Manual alignments are most common, but the correlation process becomes more time‐consuming and complex with increasing numbers of data sets. Automating stratigraphic correlation has the potential to make this process less user‐dependent and pave the way for a semi‐supervised machine‐learning approach. Dynamic time warping (DTW) has been applied as an automated correlation technique in various geoscientific studies. However, fully‐automated correlation performs poorly with complex data sets and cannot be trusted blindly. In this study, we apply and modify the DTW technique to automate the correlation of industrial downhole‐logging natural gamma radiation (NGR) signals from the Northwest Shelf of Australia. We present a DTW‐based algorithm that ensures stratigraphic accuracy in areas with considerable spatial variability between NGR signals. First, detrending and rescaling NGR‐logs is essential to make records with a different subsidence history comparable. Second, we propose a new recursion type ( step pattern ) that optimizes stratigraphic alignments by limiting squeezing and stretching. Third, we propose a novel way to generate a windowing function that constrains the alignment of two timeseries, based on a priori stratigraphic information. We present several case studies in which biostratigraphic datums and manually determined correlation points are incorporated to create a user‐defined windowing function. Integrating stratigraphic constraints into the automated correlation of sedimentary sequences through DTW makes these correlations more accurate and computationally less expensive. These methodological developments create opportunities to apply industrial downhole‐logging data and similar complex stratigraphic data for paleoclimate and tectonic reconstructions.