Toward Earthquake Early Warning in Nevada: Seismic Network Configuration and Warning Time Analysis Open Access

Billions of people across the globe live in areas at risk from earthquakes. Earthquake early warning (EEW) systems aim to reduce the risk people face by rapidly detecting earthquakes and sending alerts to people before strong shaking arrives at their location (e.g., Allen and Melgar, 2019; Wald, 2020; Allen and Stogaitis, 2022). The United States has an EEW system in place called ShakeAlert that aims to detect earthquakes within or near California, Oregon, and Washington and issue alerts to residents of those states (e.g., Given et al., 2014, 2018; Kohler et al., 2018, 2020; Minson et al., 2022; Lux et al., 2024; McGuire et al., 2025). The neighboring state of Nevada has comparable earthquake hazard to these three states (Slemmons et al., 1965; Ryall, 1977; Anderson and Miyata, 2006; Anderson et al., 2019), with the unfortunate coincidence of high population density in the areas with the greatest earthquake hazard (Fig. 1). According to the 2023 report from the Federal Emergency Management Agency on estimated earthquake risk and loss, Nevada has an estimated annualized earthquake loss (AEL) of almost 300 million dollars and is in the list of top ten states with the largest populations exposed to dangerous ground shaking (Jaiswal et al., 2023). The report also highlights Reno and Carson City as being very at‐risk areas with high AEL and annualized earthquake loss ratios per capita similar to those in Los Angeles and San Francisco, California, respectively (Jaiswal et al., 2023). The state also hosts critical infrastructure including various military bases, the Nevada National Security Site, and numerous mining and geothermal energy operators. Bringing EEW to the state could greatly reduce that risk that residents of Nevada face and also improve ShakeAlert performance in eastern California, where recent earthquake sequences have proven challenging for the current system due to limited station coverage (Lux et al., 2024).

One of the most important factors controlling the speed and accuracy of EEW alerts is the station density (Kuyuk and Allen, 2013b; Allen and Melgar, 2019; Kohler et al., 2020; Wald, 2020; Böse, Papadopoulos, et al., 2022). Typically, the more stations near a source, and the better surrounded it is, the faster and more accurately the event is detected and alerts can be sent out. Increasing the station density also reduces the size of the late‐alert zone, the area close to the source where it is not possible to issue an alert before the shaking arrives (Kuyuk and Allen, 2013b; Allen and Melgar, 2019). Increasing the number of stations in Nevada could likewise help improve alerts for events near the state border and benefit neighboring states like California where ShakeAlert is currently active (Kuyuk and Allen, 2013b; Lux et al., 2024).

An important step to facilitate a potential expansion of ShakeAlert to Nevada is analyzing the effectiveness of the current seismic network geometry and the influence of increasing station density. Doing so provides a baseline for expected algorithm performance and allows us to quantify the extent to which the installation of new stations would improve ShakeAlert’s performance in the state. ShakeAlert uses the existing infrastructure of the Advanced National Seismic System for its real‐time data streams, including some but not all stations operated by regional seismic networks. At present, ShakeAlert only uses a small fraction of stations (25 in total) operated by the Nevada Seismological Laboratory (NSL), mostly along the California‐Nevada border. Our study involves the creation and analysis of 34 representative earthquake scenarios, comparing the effectiveness of the ShakeAlert system with the network as currently formed with the addition of the other 108 available qualifying stations operated by the NSL. We also tested the effects of varying the station density by adding regular grids of potential new stations to the combined ShakeAlert and NSL‐operated network. These analyses demonstrate the potential of the current NSL‐operated network both for implementation of EEW in Nevada and for improving warning times to neighboring states, including California.

In an ideal world, we would be able to always meet our desired station density and add new stations anywhere. But in practice, equipment installation and permitting resources are always limited, and thus it is useful to identify the best locations to add new stations to improve early warning performance metrics. For this purpose, we adapt the upgrade score method proposed by Hotovec‐Ellis et al. (2017) to provide quantitative guidance in accounting for multiple factors that influence algorithm performance. Upgrade scores can be used to target the best locations for adding new stations when resources are limited. In the subsequent sections, we describe the methods used to address basic questions about early warning performance in Nevada and the implications our findings may have.

The foundation for testing the effectiveness of various network configurations is calculating the warning time. Most EEW systems require multiple stations to trigger before an alert will be sent out to reduce the number of false alerts and better characterize the event size and location. ShakeAlert sets a minimum of four stations (Kohler et al., 2020) for its first alert, so we do as such here. For each possible earthquake scenario, the warning time is defined as the difference between the arrival of the P wave at the fourth closest station to the source and the arrival of the S wave at the location of interest, adding a nominal latency of 4 s to account for data transmission, processing, and alert time (e.g., Kuyuk and Allen, 2013b). Although there have not at present been dedicated studies of NSL station latencies, we anticipate rapid transmission speeds due to a dedicated and optimized microwave network. However, processing and alert latencies should be comparable to what is observed for other ShakeAlert regions, so we use the fixed 4 s value for consistency with previous work.

The 34 distinct warning time scenarios considered here were developed from (1) notable historical earthquakes in Nevada, (2) ShakeAlert detected events that occurred near the California‐Nevada border, and (3) possible event locations that are of concern or interest based on known faults that contribute to earthquake hazard in the region (Hatem et al., 2022). All events were modeled as point sources. Warning times were calculated and compared for the current ShakeAlert network, which includes some but not all qualifying stations operated by the NSL, and for a combined network including the current ShakeAlert stations alongside all 108 potential qualifying NSL additions, referred to from now on as the SA + NV network. In additional tests, to model the effects of increasing station density, we created regular grids of evenly spaced stations using intervals between 100 and 5 km, running each spacing one hundred times, shifting 10% of the spacing in latitude or longitude each time to mitigate the effect of the absolute grid position and instead isolate the effect of grid density, which is our main focus. These grids are added to our proposed SA + NV network, and the average found across all the shifts and then compared to the other densities and the original SA + NV network without any new additions.

The ShakeAlert network as of 2023 included 25 NSL stations, a fraction of Nevada’s total 108 available stations for EEW (Fig. 2). These stations are included as part of the current ShakeAlert network used in models created for the historic and potential events. Any restricted or otherwise incapable stations in the NSL network that are unlikely to be able to provide publicly available real‐time data reporting were omitted from this analysis. These include select stations used primarily for monitoring mining operations or national security experiments, as well as older stations that do not meet the instrumentation requirements for ShakeAlert.

Our methodology for calculating upgrade scores is based on the method created by Hotovec‐Ellis et al. (2017) to assess the potential for EEW in Hawaii. When calculating the upgrade score, we consider (1) the distance to nearest station, (2) the warning time improvement, and (3) the azimuthal gap. Distance to nearest station acts as an estimation of local station density. Azimuthal gap quantifies how well a source is geometrically surrounded by the four closest stations, influencing the accuracy of the initial source location and magnitude estimate. The change in warning time estimates how installation of a new station will affect warning times. Because hazard is broadly distributed in Nevada (e.g., Petersen et al., 2024), we do not weight these values by a hazard‐based metric.

Our process starts by creating a grid of points across the state, with a spacing of 0.1°, to act as both the locations for potential new stations and the point sources for potential earthquakes. Distance to nearest station is calculated for each point in the grid of potential sources. The calculations involve looping over all the possible source points for each station point, considering how a new station will change the warning time and azimuthal gap for all possible events that could occur in the state. For these parameters, which are described in more detail subsequently, we focus on how a new station will affect the earliest alerts and reports created, so a new station will only change the warning time or azimuthal gap if it is closer than any of the current four closest stations to the source.

For warning time improvement, we find how the proposed station affects the time for the P wave to arrive at the fourth closest station to a source grid point (P4 time) because this directly influences how soon an alert can be triggered and warnings sent. For every source grid point, the current P4 time is calculated. Then for every proposed station, the change in P4 time, if any, is found for every source grid point. These values are summed up to create the warning time improvement score for the proposed station.

Azimuthal gap is defined in this context as the largest angle between any two adjacent stations of the closest four stations surrounding a source, with the source point acting as the vertex between them. A symmetrically surrounded station will have an azimuthal gap of 90°, whereas an azimuthal gap of 180° or greater indicates that all four stations are distributed to one side of the source. The smaller the azimuthal gap, the better surrounded a source is, and the more accurate the source will be, characterized in terms of location and magnitude (on average). When an EEW system is actively detecting an event, more stations will add their data over time, usually decreasing the azimuthal gap and improving the location estimate. Here, we focus on the azimuthal gap of the four closest stations to see the improvements in the initial location estimate that adding a new station will cause. For every station grid point, we loop over every source grid point and determine whether the station grid point is closer to the source than any of the original four stations. If it is not, the value tracking change in azimuthal gap for that source point is set to zero for that station grid point, the same as it would be with warning time. If it is closer, than the change in azimuthal gap is also calculated. Similar to the warning time improvement scores, the azimuthal gap scores for each station grid point are aggregated across all source grid points. Change in azimuthal gap is the only parameter that can result in a negative or positive value because it is possible that the addition of a station will increase the azimuthal gap when only the four closest stations are considered.

To illustrate the method and build intuition about the underlying results, we first describe two specific event scenarios: (1) the 1915 M 7.5 Pleasant Valley earthquake, the largest historical earthquake to occur in Nevada since 1900, and (2) a hypothetical event on the most active portion of the Death Valley fault (refer back to Fig. 2 for slip rates and fault location). For each scenario, we compare expected warning times for the current ShakeAlert network (SA) with expected warning times for the combined SA + NV network (Fig. 3). In both cases, the integration of SA + NV stations improves the warning time (by 8 s for Pleasant Valley and 4.5 s for Death Valley) and reduces the radius of the late‐alert zone (by 28.0 km for Pleasant Valley and 14.8 km for Death Valley). The Pleasant Valley scenario shows larger improvements because the Death Valley region has denser SA coverage than the Pleasant Valley region.

Lux et al (2024) discuss some of the challenges that the ShakeAlert system encountered in its first five years of operation (October 2019 through September 2023). One of their main findings is that missed or inaccurate alerts are relatively common for events located at the edges of the alerting region, which have lower station density and larger azimuthal gaps. This was the case for two California‐Nevada border events: the 2021 M 4.7 Truckee earthquake and the 2021 M 6.0 Antelope Valley earthquake. At the time of these events, NSL stations were not used to determine ShakeAlert solutions.

Warning time scenarios for the Truckee and Antelope Valley earthquakes are shown in Figure 4, comparing the difference between the SA network of the time, which did not include any NSL stations, and the proposed SA + NV network. In both cases, the warning time improves by almost five seconds for most of the state and the late‐alert zone is reduced: from 40.6 to 22.3 km radius for Truckee and from 46.3 to 30.6 km radius for Antelope Valley. The addition of the NV stations also improves the azimuthal gaps around these events, better surrounding them, which would have most likely improved the location and magnitude estimates for these events.

Nevada has high hazards and active faults near some of its most densely populated areas. We modeled the worst‐case scenarios, in terms of warning time, for the cities of Reno–Carson and Las Vegas by modeling events on nearby faults of concern to each city (Fig. 5): the Reno–Carson scenario on the Mount Rose fault system and Las Vegas scenario on the Frenchman Mountain fault system (e.g., Anderson et al., 2019; Eckert et al., 2021). Improvements in warning time are larger (up to 16 s compared to 3 s) for the Frenchman Mountain scenario near Las Vegas, which has minimal coverage by SA alone but dozens of NV stations nearby. It is important to note that for both of these scenarios, the proximity of the sources to the cities means that a large population will remain in the late‐alert zone even with enhanced station coverage. Even though warning times are improved and the late‐alert zone reduced, dramatically so for Las Vegas, the worst‐case scenarios would still provide very little or no warning times for much of the exposed area. However, an increased station density may provide a more accurate ShakeAlert epicenter and magnitude estimate, a consideration we neglect in the warning time analysis.

These examples illustrate that interactions between source location and station geometry combine to control EEW performance metrics. To comprehensively test how increasing the station density affects warning times, we consider a suite of 34 earthquake scenarios from historical events and potential sources along active faults in Nevada. The scenarios were run adding regular station grids of varying density to the combined SA + NV network. Grids with station spacing ranging from 100 to 5 km were tested. Results for the Pleasant Valley scenario (Fig. 6) demonstrate that adding even a coarsely spaced grid of 100 km spacing improves warning time by 13 s, a notable improvement for only relatively few stations added near the epicenter. With each increase in grid density, the warning time continues to improve. However, the marginal improvement of adding more stations eventually diminishes, with 10 and 5 km spacing for this event producing near identical improvements of 23 s.

The findings for the Pleasant Valley scenario are validated by the statistics aggregated across all 34 scenario events. In Figure 7, we show the average warning time improvement for areas within 100 km of the source as well as the late‐alert zone radius as a function of grid density. The absolute values of the warning times improvements and late‐alert zone radii vary between scenarios depending on the source location and its placement within the SA + NV network. However, the dependence of these metrics on station grid density is quite consistent. Every event shows increases in expected warning times as the station density increases. Similarly, the radius of the late‐alert zone decreases with increasing station density, though due to processing, transmission, and alert latencies (here assumed to be 4 s), it is impossible to achieve a radius of 0 km even for very dense networks because processing and transmission delays will always define an area that does not receive an alert before shaking arrives.

To summarize the potential of the SA + NV network outside of our designated scenario locations, we compare the first alert times of the SA and SA + NV networks for all potential source locations across the state (Fig. 8). With the current SA network (Fig. 8a), events occurring throughout most of Nevada would not trigger alerts until at least 20 s or longer after the event has started. Including all potential Nevada stations (SA + NV, Fig. 8b) improves first alert times across the state and, notably, there are more areas with first alert times less than 5 s, particularly at densely populated areas.

While this article was in preparation, on 9 December 2024, the $Mw$ 5.7 Parker Butte earthquake occurred near Yerington, Nevada, east of Reno and Carson City. Shaking was widely felt throughout Reno and Carson City (MMI IV), and EEW alerts through the ShakeAlert system, including via the Wireless Emergency Alert system, were distributed to people across the border in California. Despite being on the edge of its present network, ShakeAlert performed well for this event, with an initial magnitude estimate of M 5.7 and location error of 3 km. If ShakeAlert was actively releasing warnings to Nevada residents at the time of the earthquakes, our calculations show (Fig. 9) that the integrated SA + NV network could have provided Reno with 5–10 s of warning time. Increasing station density in the near‐source region could have provided up to an additional ten seconds of warning time.

Figure 10 displays an upgrade score map for Nevada, along with the individual factors that contribute to its calculation. Areas with higher upgrade scores indicate locations that would most benefit the EEW system with additional seismic stations. Recall from equation (1) that the upgrade score is composed as an unweighted sum of three terms: DTNS, AZGS, and WTIS. All three terms are normalized before the final upgrade score calculation, and are shown as such in the figure, except for DTNS, which is displayed in kilometers to be more understandable. DTNS acts as a representation for local station density, with lower station densities associated with higher DTNS values. All else equal, a location with higher DTNS values would be a better location to place new stations, filling in areas with less coverage. Much of central and eastern Nevada meets this criterion.

To accurately estimate shaking with sufficient warning time, we need to rapidly identify the location and magnitude of the event source. In general, the magnitude and location estimates will tend to be more accurate with good azimuthal station coverage, so minimizing azimuthal gaps in the network is an important consideration for EEW. Figure 10b shows the overall affects a new station would have on azimuthal gap, with some locations being worse than others because they would increase the gap. The areas with the largest negative values are where the greatest desired change would occur, and in this case that is the northeastern part of Nevada.

The main goal of EEW is to provide actionable warning times, so it is important that this information is incorporated into the upgrade scores. Figure 10c shows the results for warning time improvement for Nevada. Our calculations consider the actual change in warning times that adding a station would cause. The goal is to identify locations that cause the largest improvements to warning times. In this case, the area with the highest score, where adding a station would cause the most improvement, is in central Nevada and extending eastward to the Utah border.

The final upgrade score is derived by combining together these three components. The final upgrade scores shown in Figure 10d come from weighting all (self‐normalized) components equally. The highest scores for the whole state are located in the northeastern corner of the state, where there are no NSL‐operated stations delivering public, real‐time data. Although forecasted hazard is lower than in western Nevada, large earthquakes regularly occur in the eastern part of the state, including the 2008 M 6.0 Wells earthquake (Smith et al., 2011). However, because it is possible that a future ShakeAlert expansion to Nevada would focus, at least initially, on the more populated areas of the western half of the state, zooming in on these target regions (Fig. 11), additional stations are most needed to the east of these population centers, where station coverage is presently sparse. The region east of Reno has hosted the largest earthquakes in Nevada within the historic record (e.g., the 1915 M 7.3 Pleasant Valley and 1932 M 7.2 Cedar Mountain earthquakes, as well as the recent 2024 M 5.7 Parker Butte event; Fig. 9).

To demonstrate a practical use of upgrade scores, we iteratively identify the ten best locations to put new stations for Nevada. Working with the current upgrade score (Fig. 12a), we find the location with the highest value within state borders and add one station there. The upgrade score is then recalculated including the new station, and we find the new location with the highest value (Fig. 12b). This process is infinitely repeatable, allowing for the addition of any number of stations (results for five and ten stations are shown in Fig. 12c,d). Even with just ten added stations, we see first alert times decreasing by up to 30 s across the state.

As presently configured, the NSL network is concentrated along the California‐Nevada border, but this study illustrates that other areas would benefit from additional stations. The central Nevada seismic belt has a known history of large events, including five M 7+ events since 1840 (Slemmons et al., 1965; Ryall, 1977), and eastern Nevada hosted an M 6.0 earthquake near Wells in 2008 (Smith et al., 2011). Despite the clear hazard, there are very few stations near these potential sources. Closer to Reno and Carson City, the 2024 M 5.7 Parker Butte earthquake (Fig. 9) demonstrated the need for better azimuthal coverage to the east of these cities. Overall, EEW system performance would benefit greatly from adding more stations to these areas, as demonstrated by the grid density scenarios (Fig. 7) and the upgrade score results (Fig. 10) considered in this study. An important note is that we only consider here the integration of stations operated by the NSL; adding stations from University of Utah Seismograph Stations alongside NSL stations would change the results, mainly along the eastern border, likely lowering the upgrade scores there and reducing the number of stations needing to be added to reach an adequate station density.

It is notable that, for the grid scenarios, the mean warning time improvements vary by less than a factor of two from a coarse spacing of 50 km (∼3 s) to a fine spacing of 5 km (∼7 s), which implies that substantial improvements can be achieved even with a small number of station installations. However, reducing the late‐alert zone radius is critical for scenarios in which population centers are near the source, so increasing the station density as much as possible can be beneficial, especially in highly populated regions. Higher station densities also come at a higher cost (likely $50,000–$100,000 per site in Nevada, e.g., Given et al., 2018), so we must balance a desire for accuracy and speed with the limited resources available to expand the network. For a given area where higher station density is needed, it takes four times as many stations to achieve 10 km spacing compared to 20 km, and it takes 16 times as many stations to achieve 5 km spacing. A target spacing of 20 km matches well with the recommendations of the 2018 Revised ShakeAlert Technical Implementation plan (Given et al., 2018), which recommends 20 km near potentially hazardous seismic sources, 40 km in low‐risk areas, and 10 km in highly populated areas. The station density also does not need to be uniform across the whole state. Focusing on areas with a known history of large events or with high hazard or active faults might provide the most benefit. Using the upgrade score for these more focused areas (Fig. 11) could help locate the ideal location to put new stations to help make up for any areas where it is not feasible to achieve the ideal station density.

The results of the upgrade score analysis indicate that the area with the highest score is the northeast corner of the state. Although this result might seem obvious just by looking at the station map, the upgrade score method can be further tailored or used in other locations. It is possible to more heavily weight one parameter over the other if, for example, we would prefer to maximize warning time or focus on improving the azimuthal gap to try to have more accurate locations, and the method can also be modified to also incorporate hazard or risk weighting. And as shown in Figure 12, repeated iterations of the calculations can be done to best place any number of stations within a target geographic area. The main drawback of this method as presently implemented is that it only considers adding one station at a time to the network when in reality multiple stations are likely to be added at the same time. For these cases, a revised upgrade score could be calculated in which multiple station additions are considered simultaneously. In addition, the improvement metrics used to define the upgrade score emphasize the first alert, neglecting updates to alert performance because more distant stations are triggered and contribute to the solution.

Including additional Nevada stations is just the first step toward fully integrating Nevada into ShakeAlert. The essential aspect of ShakeAlert and most EEW systems is estimating the location, magnitude, and ground motion of an earthquake once it has started. ShakeAlert has multiple methods of doing this working together to provided redundancy and cover for any issues that any one method has (Böse et al., 2012, 2023; Böse, Andrews, et al., 2022; Given et al., 2018; McGuire et al., 2025) and avoid saturation for large‐magnitude events (e.g., Trugman et al., 2019). These methods, including the EPIC and FinDer codes, need to be tested in Nevada to make sure they will work here. EPIC uses a global relation to estimate magnitude from peak P‐wave displacement (Kuyuk and Allen, 2013a; Chung et al., 2019), and we have already begun testing the effectiveness of that relation with Nevada events. FinDer (Böse et al., 2012, 2023) heavily relies on having a more regular station network surrounding any possible sources. Nevada’s network is much more irregular, though, compared to California, so it is unclear how effective FinDer or the station‐centric Propagation of Locally Undamped Motion (e.g., Kodera et al., 2018; Kilb et al., 2021; Saunders et al., 2022) algorithms will be without additional dedicated studies. ShakeAlert also now integrates real‐time geodetic data in creating alerts for large events (Murray et al., 2018, 2023; McGuire et al., 2025). At present, there are few geodetic instruments equipped for real‐time monitoring in Nevada. More research would be needed to understand their value for future Nevada earthquakes, which could include large normal‐faulting events that are not commonly observed in current ShakeAlert states.

We perform a systematic assessment of seismic network configurations and event scenarios in Nevada with an aim to understand the effective potential of EEW for the state. We show that the present Nevada network configuration can already provide useful warning times, and further expansion could lead to important improvements in warning times and reductions in late‐alert zone areas. The full integration of Nevada stations into ShakeAlert would also improve warning times to California and neighboring states for events near the border. Increasing the station density to a target value as sparse as 50 km could improve warning times by an average of 3 s across scenarios, whereas a more ambitious but still tractable goal of 20 km spacing could improve warning times by about 5 s. The upgrade scores provide a framework to help compare proposed sites and better locate sites to put new stations, identifying the area most in need of stations as the northeast corner of Nevada. If a future ShakeAlert expansion focuses on the western half of Nevada, then the areas just east of Reno, Carson City, and Las Vegas would be prioritized by this metric.

The list of active ShakeAlert stations was provided to us through a personal communication with J. Saunders. Station information and metadata for Nevada Seismological Laboratory operated stations is submitted to and publicly archived by the Earthscope’s Incorporated Research Institutions for Seismology Data Management Center (IRIS‐DMC). The population data are from the Center For International Earth Science Information Network‐CIESIN‐Columbia University (2018) The fault database used to define earthquake scenarios is from Hatem et al. (2022). Quaternary faults and folds used for visualization in Figure 1 were obtained from https://www.usgs.gov/programs/earthquake-hazards/faults#data (last accessed March 2025).

The authors acknowledge that there are no conflicts of interest related to this work.

The authors thank K. Bogolub and W. Savran for interesting scientific discussions and for help assembling information about seismic network data in Nevada, and D. Kilb and J. Saunders for their thoughtful and constructive reviews that greatly enhanced the article. This material is based upon work supported by the U.S. Geological Survey under Grant Number G23AP00212 and Nevada Division of Emergency Management Hazard Mitigation Grant Program Award Number HMGP DR‐4523‐08‐08P. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the opinions or policies of the U.S. Geological Survey. Mention of trade names or commercial products does not constitute their endorsement by the U.S. Geological Survey.