Small-magnitude earthquakes shed light on the spatial and magnitude distribution of natural seismicity, as well as its rate and occurrence, especially in stable continental regions where natural seismicity remains difficult to explain under slow strain-rate conditions. However, capturing them in catalogs is strongly hindered by signal-to-noise ratio issues, resulting in high rates of false and man-made events also being detected. Accurate and robust discrimination of these events is critical for optimally detecting small earthquakes. This requires uncovering recurrent salient features that can rapidly distinguish first false events from real events, then earthquakes from man-made events (mainly quarry blasts), despite high signal variability and noise content. In this study, we combined the complementary strengths of human and interpretable rule-based machine-learning algorithms for solving this classification problem. We used human expert knowledge to co-create two reliable machine-learning classifiers through human-assisted selection of classification features and review of events with uncertain classifier predictions. The two classifiers are integrated into the SeisComP3 operational monitoring system. The first one discards false events from the set of events obtained with a low short-term average/long-term average threshold; the second one labels the remaining events as either earthquakes or quarry blasts. When run in an operational setting, the first classifier correctly detected more than 99% of false events and just over 93% of earthquakes; the second classifier correctly labeled 95% of quarry blasts and 96% of earthquakes. After a manual review of the second classifier low-confidence outputs, the final catalog contained fewer than 2% of misclassified events. These results confirm that machine learning strengthens the quality of earthquake catalogs and that the performance of machine-learning classifiers can be improved through human expertise. Our study promotes a broader implication of hybrid intelligence monitoring within seismological observatories.

The scientific literacy level of the whole population has long been focusing the researchers’ attention because of its direct impact on many aspects of our lives. As a matter of fact, studies in cognition have both been inspired by educational issues as well as by misconceptions of scientific ideas often based on irrational beliefs, old theories, unscientific reasoning, or unassimilated conceptual instruction. As a result, individual conceptions are now accurately described in many scientific fields, which has led to improvements in science teaching and learning. However, the community (scientists, academics, high school and primary school teachers, and educators) has not yet succeeded in solving all the issues, so some pre-existing misconceptions still persist in the population. In this paper, we argue that cognition studies must now focus on the origin of individuals’ conceptions and on their modes of acquisition and propagation. The goal is to provide educational tools for acting upstream, during early scientific instruction, before the very acquisition of scientific conceptions.

<p>Due to the complexity and high dimensionality of seismic catalogues, the dimensional reduction of raw seismic data and the feature selections needed to decluster these catalogues into crisis and non-crisis events remain a challenge. To address this problem, we propose a two-level analysis.</p> <p> </p> <p>First, an unsupervised approach based on an artificial neural network called self-organising map (SOM) is applied. The SOM is a machine learning model that performs a non-linear mapping of large input spaces into a two-dimensional grid, which preserves the topological and metric relationships of the data. It therefore facilitates visualisation and interpretation of the results obtained. Then, agglomerative clustering is used to classify the different clusters obtained by the SOM method as containing background events, aftershocks and/or swarms. To estimate the classification uncertainty and confidence level of our declustering results, we developed a probabilistic function based on the feature representation learned by the SOM (spatiotemporal distances between events, magnitude variations and event density).</p> <p> </p> <p>We tested the two-level analysis on synthetic data and applied it to real data: three seismic catalogues (Corithn Rift, Taiwan and Central Italy) that differ in area size, tectonic regime, magnitude of completeness, duration and detection methods. We show that our unsupervised machine learning approach can accurately distinguish between crisis and non-crisis events without the need for preliminary assumptions and that it is applicable to catalogues of various sizes in time and space without threshold selection.</p>

Many applications in seismology require to isolate earthquake clusters from a background activity. Relative declustering methods essentially find a 2D representation of an earthquake catalogue that distinguishes between two classes of events: crisis and non-crisis events. However, the number of statistical and/or physical parameters to be used is often limited due to the difficulty of concatenating the information onto a physically meaningful 2D grid. In this study, we propose to alleviate the declustering task by using the ability of unsupervised artificial intelligence to model complex spatio-temporal relationships directly from data. Through a data-driven approach, we define an easily transferable declustering model that provides declustering results with fewer assumptions and no prior selection of thresholds. We first obtain this model by training a self-organising neural network (SOM) that learns to cluster data points according to their feature similarity on a 2D map. We then assign each SOM cluster a label (crisis or non-crisis class) using an agglomerative clustering procedure. We quantify the classification uncertainty by developing a probabilistic function based on the projection learned by SOM. Our method is applied to a synthetic dataset and to real catalogues from the Gulf of Corinth, Central Italy and Taiwan. We discuss the validity of the method by estimating its classification accuracy. For real data, we qualitatively compare our results to previous declustering attempts. We show that our approach is easy to handle, provides a fairly new representation of earthquake catalogues and has the potential to reduce classification ambiguities between nearby events.

Many applications in seismology require to isolate earthquake clusters from a background activity. Relative declustering methods essentially find a 2D representation of an earthquake catalogue that distinguishes between two classes of events: crisis and non-crisis events. However, the number of statistical and/or physical parameters to be used is often limited due to the difficulty of concatenating the information onto a physically meaningful 2D grid. In this study, we propose to alleviate the declustering task by using the ability of unsupervised artificial intelligence to model complex spatio-temporal relationships directly from data. Through a data-driven approach, we define an easily transferable declustering model that provides declustering results with fewer assumptions and no prior selection of thresholds. We first obtain this model by training a self-organising neural network (SOM) that learns to cluster data points according to their feature similarity on a 2D map. We then assign each SOM cluster a label (crisis or non-crisis class) using an agglomerative clustering procedure. We quantify the classification uncertainty by developing a probabilistic function based on the projection learned by SOM. Our method is applied to a synthetic dataset and to real catalogues from the Gulf of Corinth, Central Italy and Taiwan. We discuss the validity of the method by estimating its classification accuracy. For real data, we qualitatively compare our results to previous declustering attempts. We show that our approach is easy to handle, provides a fairly new representation of earthquake catalogues and has the potential to reduce classification ambiguities between nearby events.