Using Machine Learning to Discern Eruption in Noisy Environments: A Case Study Using CO2-Driven Cold-Water Geyser in Chimayo, New Mexico

We present an approach based on machine learning (ML) to distinguish eruption and precursory signals of Chimayo geyser (New Mexico, U.S.A.) under noisy environmental conditions. This geyser can be considered a natural analog of CO2 intrusion into shallow water aquifers. By studying this geyser, we can understand upwelling of CO2-rich fluids from depth, which has relevance to leak monitoring in a CO2 sequestration project. ML methods such as random forests (RFs) are known to be robust multiclass classifiers and perform well under unfavorable, noisy conditions. However, the extent of the RF method's accuracy is poorly understood for this CO2-driven geysering application. The current study aims to quantify the performance of RF classifiers to discern the geyser state. Toward this goal, we first present the data collected from the seismometer that is installed near the Chimayo geyser. The seismic signals collected at this site contain different types of noises such as daily temperature variations, animal movement near the geyser, and human activity. First, we filter the signals from these noises by combining the Butterworth high-pass (BH) filter and an autoregressive (AR) method in a multilevel fashion. We show that by combining these filtering techniques in a hierarchical fashion leads to a reduction in noise in the seismic data without removing the precursors and eruption event signals. We then use RF on the filtered data to classify the state of geyser into three classes: remnant noise, precursor, and eruption states. RF classifier is constructed based on the comprehensive features extracted using the Tsfresh Python package. We show that the classification accuracy using RF on the filtered data is greater than 90%. We also evaluate the accuracy of other classical time-series methods such as dynamic time warping (DTW) on filtered data along with RF on partially filtered data in which we remove the daily trends. Classification accuracy shows that DTW performs poorly (44%) and RF on partially filtered data performs decently 87%). Denoising seismic signals from both daily trends and human activity enhances RF classifier performance by 7%. These aspects make the proposed ML framework attractive for event discrimination and signal enhancement under noisy conditions, with strong potential for application to monitor leaks in CO2 sequestration.