Seismic Characterization of Lahars on Volcán de Fuego Toward the Development of a Machine Learning‐Based Detection Algorithm

Abstract Lahars are among the most frequent hazards associated with Volcán de Fuego, Guatemala. Despite their recurrence, early detection and automated alerts remain challenging since they often rely on manual monitoring and sparse visual confirmation. Yet, we can harness the high number of flows triggered every rainy season to characterize their seismic signatures and quantify their size and behavior. For this, we used seismic stations located along two active lahar channels on Fuego where this characterization describes a somewhat stable long‐term flow behavior. This work revealed more varied short‐term behavior characterized by increasing seismic activity in the time domain and a shift toward lower frequencies as these flows propagate downstream. Building on this characterization, we implemented K‐nearest neighbor (KNN) based detectors using seismic signal attributes describing samples of the data in the time and frequency domains, as well as statistical functions of these samples. We trained generalized and station‐specific detectors that achieved high accuracy for detecting moderate‐to‐large flows with lower performance for smaller or ambiguous events. We found that root mean square amplitude, a proxy for flow size, appears to control detector performance more than other signal features. The detector is computationally efficient and, in the case of Fuego, did not require additional instrumentation. This framework presents a portable solution for enhancing automated lahar detection while minimizing the use of location‐specific parameters required by other methods.