The starting point of the analysis is a distinction that sounds obvious but is routinely blurred in practice: the difference between a prediction and a forecast. In the authors’ terminology, a prediction is a deterministic statement that a specific hazard intensity will or will not occur in a defined region and time window, while a forecast assigns a probability to that occurrence. Even predictions, the authors note, implicitly carry uncertainty because of unavoidable false alarms and missed events, which makes them probabilistic whether scientists admit it or not. The task force argues that issuing forecasts is preferable because probabilities clarify the division of labor in risk management: scientists provide probabilities that describe scientific uncertainty, while decision-makers choose the probability thresholds that trigger action. This hazard-risk separation principle, advocated by Jordan and colleagues more than a decade ago, remains surprisingly neglected in operational practice.

Why can natural systems not be predicted deterministically in the first place? The paper points to intrinsic unpredictability arising from strong nonlinearities such as chaos, fine-scale aggregation in space and time, high dimensionality, and the open nature of Earth systems, which exchange energy and matter with their surroundings in often uncontrolled ways. On top of this natural variability sits a second, equally pervasive source of uncertainty: our limited knowledge of the underlying processes and the limitations of their numerical representation. These two categories are conventionally labeled aleatory variability and epistemic uncertainty, and the task force found that every hazard domain they examined struggles with how to define them unambiguously and how to include both in a single forecast. The proliferation of competing terms, shallow and deep, internal and external, structural and model uncertainty, likelihood and confidence, reveals fields tackling similar problems in isolation.

The technical heart of the framework lies in how forecasting models express uncertainty. Each model, whether physics-based, machine-learning, empirical, or rooted in expert judgment, produces a forecast distribution, a survival function showing the exceedance probability for each hazard intensity value. When knowledge is strong and data abundant, a single forecast distribution may suffice; the paper cites ocean biogeochemical modeling, where point estimates are compared with sparse observations to derive error statistics, and the Bluecat approach in hydrology, which converts deterministic predictions into stochastic forecasts. But in many domains, from long-term flood frequency analysis to seismic hazard assessment, no single trustable model exists. Scientists then work with an ensemble of alternative forecast distributions, each reflecting a different working hypothesis or numerical approximation. The standard practice of collapsing these into one mixture distribution, often loosely called the mean, is where the authors see a critical loss of information.

Collapsing the ensemble, they argue, discards the dispersion of the individual distributions around the mixture, and that dispersion is precisely the measure of how much scientists believe in their own assessment. A statement that an event has a 15 percent probability conveys nothing about whether it comes from a large dataset, from models in close agreement, or from averaging wildly divergent assessments. The same mixture distribution can emerge from a tightly clustered ensemble or from one spread across implausibly wide alternatives, and stakeholders deserve to know which. The proposed solution is the complete forecast: a forecast that explicitly represents and keeps distinct all recognized uncertainties. Quantitatively, it is built from what the authors call the extended experts’ distribution, a probability distribution over the exceedance probability itself, obtained by fitting the ensemble of model-derived probabilities, for example with a Beta distribution. Its spread mimics the epistemic uncertainty; its center tracks the aleatory variability.

This construction rests on a unified probabilistic framework developed by Marzocchi and Thomas Jordan, which represents both objective frequency-based information and subjective expert judgment as a distribution of probability rather than a single number. The framework is anchored in the concept of an experimental concept, a collection of observations judged to be stochastically exchangeable, whose long-run frequency defines the true, unknown forecast distribution that every model attempts to estimate. The authors are candid about the philosophical difficulties: what appears to be irreducible randomness may simply reflect incomplete understanding, and as knowledge improves, apparent aleatory variability can be reclassified as epistemic. Their resolution is to define aleatory variability not as a property of the physical process but of the data-generating process specified by the experimental concept, a definition that is model-independent in principle. If the true probability falls outside the extended experts’ distribution, the framework registers an ontological error, an unknown unknown in Donald Rumsfeld’s memorable phrase.

Testing forecasts against reality is the second pillar of the framework, and here the distinctions become operationally consequential. A single forecast distribution can be calibrated by comparing it with independent observations, for instance through the probability integral transform, which should yield uniformly distributed values for a well-calibrated model. A complete forecast, by contrast, is validated by testing whether observed exceedance frequencies are coherent with the extended experts’ distribution. When observations were used to build the model, only the humbler term consistency applies, and the authors stress that a flawed distribution can pass a consistency test through overfitting while failing calibration; only rejection is truly informative in that case. Relative skill among competing models is measured with proper scoring rules, which rank models without implying reliability. Where data are scarce or absent, as is common for the highest and most societally relevant hazard intensities, expert judgment and structured elicitation protocols take on a fundamental role in assessing whether the ensemble spans scientifically plausible interpretations.

Communication emerges as perhaps the least mature component of the entire risk management cycle. The RETURN project interviewed 32 professionals involved in risk communication across municipalities, civil protection agencies, research institutions, and consultancy, and the interviews revealed large differences in terminology and skillsets that suggest a pioneering phase rather than an established discipline. The challenges are formidable: low probabilistic literacy among the public, a tendency to mistake detail for precision, deterministic expectations of science, and technical terms such as uncertainty, error, and moderate alert that shed their precise meanings outside the laboratory. Simplifications like traffic-light color schemes feel intuitive but can produce sharp, misleading discontinuities between assessments just above or below a threshold. The task force identifies two complementary approaches, an operational one built on digestible data and a narrative one built on storytelling, noting that narratives can make tiny probabilities tangible but risk sliding into persuasion.

The framework’s implications for decision-making are illustrated by tsunami warning systems in the North-East Atlantic and Mediterranean region, where traditional decision matrices leave no room for quantifying uncertainty. A probabilistic system that links alert levels to percentiles of a forecast distribution can make explicit the trade-offs between false alarms and missed alerts, choices that are ultimately political judgments about acceptable risk rather than purely scientific ones. The authors also recommend pre-defined protocols as a transparent audit trail for emergencies, when there is no time to weigh every uncertainty but ignoring them can erode public trust. Their vision, spanning eighteen Italian universities and research centers alongside the Civil Protection Department, State Railways, and insurers, is a unified cross-hazard framework in which probabilistic forecasting serves as the common language bridging science and society, complete forecasts keep epistemic and aleatory uncertainties distinct, rigorous evaluation establishes credibility, and communication experts ensure the last mile to decision-makers is never left uncovered.

Subject of Research: A cross-hazard framework for quantifying and communicating uncertainty in natural hazard forecasting

Article Title: Invited perspective: Uncertainties in natural systems may be uncomfortable, but ignoring them would be absurd

Article References: Marzocchi, W., Montanari, A., & the RETURN-uncertainty task force (2026). Invited perspective: Uncertainties in natural systems may be uncomfortable, but ignoring them would be absurd. Natural Hazards and Earth System Sciences, 26(10), 4861-4880. https://doi.org/10.5194/nhess-26-4861-2026

Image Credits: AI Generated

DOI: 10.5194/nhess-26-4861-2026

Keywords: uncertainty, natural hazards, probabilistic forecasting, epistemic uncertainty, aleatory variability, risk communication, complete forecast, seismic hazard, flood forecasting, model evaluation, decision-making, RETURN project

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