Automatic detection and classification of marine and fluvial terraces using statistical and stochastic clustering methods

Terrace landforms, including marine, fluvial, and lacustrine terraces, play a significant role in various geoscience fields as records of past relative water-level changes caused by climate and tectonic activity. The identification of lateral continuity of synchronous terraces is one of the most important observations. However, the evaluation of terrace continuity often encounters difficulties due to erosion and weathering, and often relies on subjective judgment. This study improves upon the previous terrace clustering method by applying stochastic analysis using a Gaussian Mixture Model, achieving an automatic and highly reliable classification. For model verification, we applied this classification analysis to the Pleistocene marine terraces in the Huon Peninsula, Papua New Guinea; the Holocene marine terraces in the Boso Peninsula, Japan; and fluvial terraces along the Waipawa River, New Zealand; all of these cases have been well studied in previous research and have high-resolution terrain data available. In each study area, a quantitative and graphical representation of the continuity and likelihood of cliff features is provided. The classification process is implemented with a Python script and is able to semi-automatically detect and classify terraces. The wide adaptability, easy application, and quick implementation of this model, accompanied by the recent expansion of worldwide topographic datasets due to advancements in remote sensing, will accelerate the analysis of global terraces. • Developed a numerical model for automatic detection of terrace landforms • Detected lateral continuity of terraces using stochastic clustering methods • Improved objectivity compared to conventional numerical approaches • Demonstrated high applicability to diverse terraces worldwide • Released as open-source Python code, making it easily accessible and usable