The work was led by Daniel Martinez-Jaramillo, a doctoral researcher at the Universidad Nacional Autónoma de México, together with Sreeram-Reddy Kotha of ISTerre in Grenoble, F. Ramón Zúñiga, and Pierre Lacan. Their starting point was the national seismic network of the Colombian Geological Survey, known by its FDSN code CM, which has operated since 1993 and currently comprises 506 seismological and accelerometric stations distributed across the country. Local networks monitoring volcanoes, mining districts, and oil and gas fields contributed additional stations. Upon request, the survey supplied more than 10,000 quality-checked acceleration time series. The team applied strict selection criteria: events between 0 and 14 degrees north latitude and 69 to 82 degrees west longitude, at depths shallower than 50 kilometres, with magnitudes above 4 and stations within 400 kilometres of the epicentre. After discarding volcanic signals and spurious spikes, 667 earthquakes and 7550 records survived the cut, forming what seismologists call a flatfile: a single table describing every earthquake, every recording site, and every measured level of shaking.

The technical care invested in processing these waveforms is what elevates the dataset from a raw archive to a scientific instrument. Each acceleration time series underwent the same sequence of corrections, following protocols established for the Italian strong-motion databases: baseline correction, cosine tapering, a second-order acausal bandpass Butterworth filter, double integration to obtain displacement, linear detrending of that displacement, and double differentiation back to corrected acceleration. The high-pass corner of the filter was made magnitude-dependent, ranging from 0.15 hertz for the smallest events down to 0.05 hertz for the largest, while the low-pass corner was fixed at a median value of 32.26 hertz. A sensitivity analysis showed that this fixed low-pass choice barely affects most records, shifting the orientation-independent RotD50 peak ground acceleration by an average of just 2.31 percent, though 6.4 percent of recordings, concentrated around magnitude 4.5 at distances beyond 250 kilometres, showed variations above 10 percent, likely reflecting high-frequency noise near the filter corner.

The intensity measures extracted from the corrected records are comprehensive. The dataset provides peak ground acceleration, velocity, and displacement for all three components, along with 5 percent-damped spectral accelerations at 31 oscillator periods spanning 0.01 to 8 seconds, computed as RotD50 values, the median over all non-redundant horizontal orientations, a measure now standard in modern ground-motion modelling. Fourier amplitude spectra for each component cover frequencies from 0.04 to 50 hertz, smoothed with the Konno-Ohmachi function, and the effective amplitude spectrum, a geometric combination of the two horizontal components, is also included. Crucially, the authors report the lowest and highest usable frequencies for every record, bounded by a safety factor of 1.25 around the filter corners so that users know exactly which spectral values are trustworthy. This usable-bandwidth information, often omitted from older compilations, allows engineers and seismologists to weight each record appropriately in downstream analyses.

Perhaps the most technically inventive element concerns magnitudes. Moment magnitude is the preferred scale for hazard analysis, but reliable agency-reported values were not available for every event. Where they were missing, the team estimated magnitudes from the recordings themselves by fitting a single-corner Brune omega-squared source model to each corrected spectrum, using a coarse grid search followed by non-linear least-squares refinement in log-log space. From the resulting corner frequency, the source radius was computed assuming a shear-wave velocity of 3.5 kilometres per second, and the seismic moment followed from the Eshelby circular-crack relation under an assumed constant stress drop. Testing stress drops of 1, 3, and 5 megapascals against independently reported moment magnitudes from the Global CMT catalogue and other agencies, the authors found that 5 megapascals reproduced catalogued values best. This procedure homogenised the entire catalogue, ultimately widening the magnitude range to 3.5 through 7.2, with 24 percent of events falling below magnitude 4 once recalculated.

Site characterisation received equal attention. The 227 recording stations are described by their time-averaged shear-wave velocity in the upper 30 metres, the widely used Vs30 parameter, together with horizontal-to-vertical spectral ratios and predominant site periods derived from seismograms. For 154 sites, these parameters come from a recent northwestern South America database, and 28 of them have in-situ, microtremor-based Vs30 measurements. The remaining stations rely on Vs30 values inferred from a topographic-slope-based map of Colombia, a pragmatic but imperfect proxy. The authors are candid about this limitation, noting that proxy-based site parameters introduce additional epistemic uncertainty that may inflate the site-to-site variability observed in their validation, and pointing to H/V spectral ratio classification schemes as a promising complement for future work. Distance metrics are equally thorough: epicentral and hypocentral distances are reported for all events, while finite-fault measures such as Joyner-Boore distance are computed for the 42 earthquakes larger than magnitude 5.5, with rupture dimensions scaled from published empirical relations.

Validation came through residual analysis, the standard test of whether a new dataset behaves consistently with established ground-motion prediction models. The team compared observed peak ground acceleration, spectral accelerations, and effective amplitude spectra against the global NGA-West2 model of Abrahamson and colleagues from 2014, its regional adaptation for northern South America published by Arteta and colleagues in 2023, and the Bayless-Abrahamson 2019 Fourier-spectrum model. The residuals, decomposed into between-event, between-site, and leftover components, showed no significant biases and followed Gaussian-like distributions confirmed by Shapiro-Wilk tests. As expected, the regional model produced lower overall variability than the global one, and the between-event trends support the internal consistency of the compilation. One instructive exception emerged: within the magnitude range of roughly 4.2 to 4.7, the regional model showed a negative trend in between-event residuals, suggesting that magnitude calibration at these moderate levels could benefit from future refinement.

The statistical footprint of the dataset reveals both its strengths and its character. About 85 percent of records lie beyond 100 kilometres from the epicentre, while only 5.3 percent come from within 50 kilometres, meaning near-source shaking remains comparatively sparse. Roughly 65 percent of the records originate from events shallower than 20 kilometres, confirming the dataset’s focus on shallow crustal seismogenic sources, with the remainder spanning depths down to about 55 kilometres. Strong events are well represented: 716 records, or 9.5 percent of the total, come from earthquakes above magnitude 5.5, including the magnitude 7.2 Pizarro earthquake of 2004 on the Pacific coast and the magnitude 6.1 San Juanito earthquake of 2023 in eastern Colombia. Epicentres of 488 events fall on continental territory and 179 offshore, and the authors deliberately leave tectonic classification, whether subduction interface, intraslab, or crustal, to users, since appropriate assignments depend on the slab models each analyst chooses.

The practical payoff could be substantial. Ground-motion datasets of this kind are the raw material for probabilistic seismic hazard assessment, the framework that underlies building codes, and for the emerging generation of partially non-ergodic ground-motion models, which replace globally averaged predictions with region-specific ones by exploiting dense, well-characterised observations. Colombian engineers designing buildings, and officials updating national hazard maps, will gain a resource calibrated to the actual faults and soils beneath their feet rather than to California or Japan. The Fourier spectra additionally open a window onto source physics, allowing researchers to constrain corner frequencies, seismic moments, and stress drops for Colombian earthquakes directly. Released openly under a Creative Commons Attribution 4.0 licence on Zenodo, with the underlying time series available through the Colombian Geological Survey’s acceleration catalogue and the FDSN network CM, the dataset invites reuse far beyond Colombia’s borders, offering a template for other seismically active but data-poor regions of the world where the next damaging earthquake is only a matter of time.

Subject of Research: A uniformly processed open dataset of ground-motion recordings from shallow earthquakes in Colombia for seismic hazard analysis

Article Title: Ground-motion dataset for shallow earthquakes in Colombia

Article References: Martinez-Jaramillo, D., Kotha, S.-R., Zúñiga, F. R., & Lacan, P. (2026). Ground-motion dataset for shallow earthquakes in Colombia. Earth System Science Data, 18(10), 7403-7415. https://doi.org/10.5194/essd-18-7403-2026

Image Credits: AI Generated

DOI: 10.5194/essd-18-7403-2026

Keywords: Colombia, ground-motion dataset, seismic hazard, earthquake engineering, strong-motion records, moment magnitude, spectral acceleration, Fourier amplitude spectrum, subduction, Vs30, flatfile, open data

The study, led by Farhad Sedaghati of Aon Impact Forecasting together with Shahram Pezeshk and Mehran Davatgari-Tafreshi of the University of Memphis, focuses on a quantity called the Effective Amplitude Spectrum, or EAS. Unlike the response spectra most commonly used in building design, which describe how an idealized oscillator reacts to shaking, the EAS is derived directly from the Fourier amplitude spectrum of the recorded motion. That gives it a direct physical interpretation: it reflects how energy is distributed across frequencies by the earthquake source, modified by the path the waves travel and by the soil conditions beneath the recording station. Engineers can also feed EAS predictions into random vibration theory to estimate response spectra quickly, without running time-consuming simulations.

To train their model, the researchers drew on the NGA-West3 database, a large and uniformly processed collection of three-component accelerograms from shallow crustal earthquakes in active tectonic regions. The raw compilation contained 175,284 records, but the team applied a stringent sequence of quality-control filters before a single neuron was trained. Records were excluded if the earthquake was smaller than magnitude 3.0, if the hypocenter was deeper than 30 kilometers, if the rupture distance exceeded 300 kilometers, or if metadata such as fault dip or shear-wave velocity were missing or implausible. Sensors buried more than two meters below the ground surface were rejected, as were stations classified as non-free-field. After all screening steps, 50,639 high-quality records remained, roughly 90 percent of them from California, with the rest contributed by Alaska, Idaho, Nevada, Utah, Washington, and Wyoming.

The neural network itself is deliberately compact. It takes six explanatory variables as input: moment magnitude, rupture distance, fault dip, depth to the top of the rupture, the time-averaged shear-wave velocity in the upper 30 meters of the site, and the depth to a stiff 2.5 kilometers-per-second shear-wave velocity horizon that characterizes sedimentary basins. From these six numbers, the network predicts the natural logarithm of EAS at 22 discrete frequencies spanning 0.05 to about 33 hertz, all at once. This multi-output design proved crucial. When the researchers initially tried predicting each frequency with a separate single-output model, the resulting spectra were jagged and unstable. Predicting all frequencies simultaneously through shared hidden layers allowed the network to learn cross-frequency patterns directly from the data, producing smooth, physically plausible spectra.

One of the thorniest practical problems was missing data. A recording may contain reliable spectral information at some frequencies but not others, because signal-to-noise ratios degrade at both the lowest and highest frequencies. Rather than discarding every record with a single gap, the team used a masked loss function during training. Missing values, represented as not-a-number entries, simply do not contribute to the error that the network tries to minimize, so partially observed spectra still teach the model whatever they can. This masked approach substantially increased the effective amount of data available, particularly at the frequency extremes where coverage is thinnest.

Avoiding inflated estimates of accuracy required equally careful design of the evaluation strategy. Recordings from the same earthquake share source characteristics, so splitting the data randomly would let the network effectively memorize events it was supposed to predict. The researchers therefore partitioned the dataset at the event level, keeping all recordings of a given earthquake together in the same fold. Hyperparameters were selected through ten-fold event-grouped cross-validation, and roughly five percent of events were held out entirely as an independent evaluation set, untouched during architecture selection, preprocessing, or training decisions. After that out-of-event assessment was complete, the final network was retrained on the full retained dataset so that every available earthquake contributed to the distributed model. The winning architecture turned out to be modest: two hidden layers with 32 and 16 neurons, a batch size of 64, and a learning rate of ten to the minus two.

The predictions that emerged show textbook seismological behavior, despite the fact that no functional form was ever imposed. Spectral amplitudes rise monotonically with magnitude and fall with distance, with high frequencies decaying faster than low ones as waves scatter and lose energy in the crust. At low frequencies the spectra steepen toward a corner frequency in a manner consistent with the classical omega-square source model, and at high frequencies they roll off in the way associated with near-surface damping. The team is careful to note that these are qualitative consistencies, not quantitative estimates of source parameters such as corner frequency or kappa, and that predictions above magnitude 7.36, the largest event in the training data, are extrapolations presented for illustration rather than validated forecasts.

Beyond median predictions, the researchers quantified aleatory variability, the irreducible randomness in ground motion that remains after source, path, and site effects are accounted for. Using a sequential mixed-effects decomposition, they separated total residuals into between-event, site-to-site, and event-site-corrected components at each of the 22 frequencies. The standard deviations show clear frequency dependence, increasing toward higher frequencies where ground motion is most sensitive to record-specific details, and they fall within the range reported by existing empirical EAS models. This decomposition matters for practice: in conventional probabilistic seismic hazard analysis the total variability is used directly, but for site-specific studies the explicit separation allows engineers to condition on measured site behavior and reduce uncertainty.

How does the machine stack up against the human-crafted equations it joins? The team compared their model against three established empirical EAS ground-motion models, Bora and colleagues from 2019, Bayless and Abrahamson from 2019, and Campbell and Bozorgnia from 2025, all built on the earlier NGA-West2 database. Across representative scenarios, the neural network’s median predictions track the empirical models closely in both spectral shape and scaling trends, and its variability estimates sit within the published range. The authors are candid that these comparisons demonstrate physical plausibility rather than superiority, since a rigorous head-to-head test would require fitting a parametric model to identical data with identical partitions, something they flag as important future work. Interestingly, the network also reproduced the observed cross-frequency correlation structure of the data with a Pearson correlation of 0.964 between observed and predicted correlation coefficients.

The researchers have released the trained model, its preprocessing scaler, and example code on GitHub, making the tool immediately usable by hazard analysts. Because the network produces broadband Fourier spectra in a fraction of a second, it is well suited to applications that demand thousands or millions of repeated ground-motion evaluations, such as probabilistic seismic hazard analysis and simulation-based risk studies. The model is strictly valid within the range of its training data, magnitudes from about 3.0 to 7.36, distances up to 300 kilometers, and sites from soft soil to hard rock, and it is calibrated specifically for western United States tectonics. Future directions include explainable artificial intelligence techniques to reveal which predictors matter most at each frequency, and physics-informed training that would embed source models such as Brune’s spectrum directly into the learning objective, potentially improving extrapolation to the largest, rarest earthquakes that engineers care about most.

Subject of Research: A data-driven neural network ground-motion model for Fourier amplitude spectra of shallow crustal earthquakes using the NGA-West3 database

Article Title: A data-driven ground-motion model for Fourier amplitude spectra of shallow crustal earthquakes using the NGA-West3 database

Article References: Sedaghati, F., Pezeshk, S., & Davatgari-Tafreshi, M. (2026). A data-driven ground-motion model for Fourier amplitude spectra of shallow crustal earthquakes using the NGA-West3 database. Bulletin of Earthquake Engineering. https://doi.org/10.1007/s10518-026-02708-1

Image Credits: AI Generated

DOI: 10.1007/s10518-026-02708-1

Keywords: ground-motion model, effective amplitude spectrum, NGA-West3, artificial neural network, machine learning, Fourier amplitude spectrum, seismic hazard analysis, shallow crustal earthquakes, aleatory variability, random vibration theory, western United States, earthquake engineering