Eruptive Regime-Shift Detection at Steamboat Geyser – Code and Data for Clustering and Monitoring Framework

This repository accompanies the conference paper "Eruptive Regime-Shift Detection at Steamboat Geyser: An Integrated ML Pipeline and Its Implications for New Zealand Hydrothermal Monitoring" (Ardid, Dempsey, Cronin). It contains all necessary data, feature sets, and scripts to reproduce the PCA–DBSCAN clustering pipeline used to detect hydrothermal regime transitions at Steamboat Geyser. Specifically, the dataset includes: Seismic data from station YNM (YNM_seismic_data.csv) Eruption metadata (YNM_eruptive_periods.txt) Feature files (2016–2022) precomputed across multiple frequency bands (rsamF, mfF, hfF, lfF, vlfF, lrarF, rmarF, vlarF, dsarF) using a 12-hour window and z-score normalization Python scripts: sensitivity_clustering_dbscan.py: performs sensitivity analysis across PCA dimensions and clustering hyperparameters, outputs visual diagnostics realtime_cluster_detection.py: simulates forward-in-time detection of new seismic regimes, mimicking a real-time monitoring system Environment file (environment.yml): includes dependencies required to run the pipeline (Python ≥3.8, scikit-learn, pandas, tsfresh, matplotlib, etc.) The methodology combines dimensionality reduction (PCA) with unsupervised density-based clustering (DBSCAN) to identify shifts in seismic behavior that precede or accompany changes in eruption dynamics. The same framework is adaptable to real-time implementation, and this dataset supports future transfer to geothermal monitoring sites in New Zealand. Usage Notes To replicate the analysis: Install dependencies using conda env create -f environment.yml Run sensitivity_clustering_dbscan.py to explore the parameter space Use realtime_cluster_detection.py to simulate sequential regime detection Plots and results are saved in the corresponding output directories Please cite the accompanying paper when using this dataset.