Gravity wave zoo: engaging citizen science to analyze atmospheric gravity wave activity over Poker Flat, Alaska

The Mesosphere and Lower Thermosphere (MLT) is a critical atmospheric region ranging from 80–140 km in altitude. A main driver of momentum transport, density perturbations, temperature variations, and background winds in this region are atmospheric gravity waves which are not well accounted for in many models of the thermosphere and near-space region. Additionally, global circulation models that do include GWs fail to resolve small-scale activity ($<$200 km). We utilize a hydroxyl (OH) airglow imager located in Poker Flat, Alaska (65°N 147°W) to leverage our active citizen science initiative, the Gravity Wave Zoo, expanding the breadth of available GW, aurora, and instability data over multiple seasons. We focus on a short-term study between 27 December 2023 and 4 February 2024 to statistically validate a subsample of this multi-year dataset, and report on recent Gravity Wave Zoo progress and accuracy, presenting a summary of overall participation and citizen classifications of GW, instability, and auroral events. We find citizen scientist classifications indicate GWs in 54.5%, aurora in 40.1%, and instabilities in 23.4% of subjects. We further propose future research directions enabled by this work and highlight the advantages of high temporal resolution data on the scale of weeks, months, and seasons.

1 Introduction

1.1 The near-space region

Beginning at approximately 80 km in altitude, the Mesosphere and Lower Thermosphere (MLT) is a dynamic and turbulent region just below the Kármán line. Within this region, complex dynamics are influenced by both the lower and middle atmosphere, as well as external drivers such as solar and space weather. These processes, however, remain difficult to predict and characterize due to observational constraints and model limitations. As a result, large-volume MLT datasets are currently scarce.

As interest grows in the managed de-orbiting of satellites and miscellaneous space debris, it becomes increasingly important to have a comprehensive understanding of regular MLT variability that occurs on time scales ranging from daily to seasonally. Densities within this region fluctuate depending on latitude, atmospheric tidal behavior, season, solar variability, and atmospheric gravity waves (GWs), the latter of which also perturb temperatures (Fritts and Alexander, 2003) and alter winds due to drag on the background mean flow (Vadas et al., 2014). This results in uncertainties and makes it difficult to accurately parameterize a vehicle’s trajectory through the atmosphere, e.g., (Pardini and Anselmo, 2009).

To address uncertainties associated with GWs, we utilize regular ground-based OH airglow imaging to produce a long-term image set of the MLT, which is analyzed with the help of citizen scientists (volunteer enthusiasts and members of the general public), filling this informational niche. The primary focus of this study is high-frequency GWs (wavelengths > 10 km and periods < 20 min), which are unresolved in global scale models of the atmosphere. A secondary focus is aurora and atmospheric instabilities (wavelengths < 10 km, with frequencies below the Brunt - Väisälä frequency). We present initial engagement from volunteers and include statistical validation of classifications. We then discuss notable data associations within our study period.

1.2 Dynamics in the mesosphere and Lower Thermosphere

1.2.1 Atmospheric gravity waves

GWs are buoyancy waves generated by the vertical displacement of an air mass. The restoring force to these oscillations is gravity, acting opposite to buoyancy (Hines, 1960; Gossard and Munk, 1954). GWs may be generated by a variety of terrestrial sources including convective storms (Alexander and Holton, 1997; Yue et al., 2009), airflow over mountainous terrain (Bossert et al., 2015), and jet stream variability and frontal systems (Becker et al., 2022; Plougonven and Zhang, 2014). GW amplitude increases exponentially with altitude and inversely with background density until a critical point is reached where dissipation occurs, resulting in a body force on the mean flow (Vadas et al., 2003). GWs are one of the key drivers of momentum transport, density perturbations, temperature variations, and background winds in the MLT (Fritts and Alexander, 2003; Hunsucker, 1982).

1.2.2 Atmospheric wave instabilities

Instabilities are dynamic phenomena associated with wind shear in the local environment and often with GW breaking or dissipation (Baumgarten and Fritts, 2014; Dong et al., 2023). These instabilities manifest as small ripples, often perpendicular to parent GWs (Hecht, 2004). Because instabilities act as signatures of GW dissipation, they are valuable targets for observation in discerning the dynamical behavior of the MLT (Fritts et al., 2022; Hecht et al., 2005).

1.2.3 Aurora

A distinct characteristic of space weather, aurora is an important marker of coupling between the neutral atmosphere, ionosphere, and magnetosphere. Aurora can be responsible for indirect transformations by altering atmospheric chemistry, including localized heating, changes in density, and the generation of GWs (Oyama and Watkins, 2012; Turunen et al., 2016). It is important to account for aurora as both a contaminant and valuable data when working with image datasets, as it can often distract from more subtle variability while providing information about the current geomagnetic environment.

1.3 Improving the breadth of Lower Thermosphere data

We provide a preliminary overview of the Gravity Wave Zoo image set–a multi-season, all-sky MLT dataset. It is our intent to classify and analyze images for high-frequency GWs, instabilities, and aurora. This is in pursuit of the ultimate goal of providing reliable, high-frequency data at longer durations, and identifying trends and variations in small-scale structure on the periods of days, weeks, months, and seasons.

1.3.1 High-frequency gravity waves and atmospheric modeling

GWs are not well accounted for in many current models of the thermosphere and near-space region. Additionally, smaller scale structures (<200 km) are not routinely resolved in global circulation models (GCMs) due to their coarse resolutions and therefore necessitate parameterizations, despite the significant role GWs play in atmospheric coupling and momentum transport (Geller et al., 2013). While recent modeling efforts aim to remedy this shortcoming, computational expenses remain high. Thus, GWs continue to be the subject of intense research efforts.

The consequences of these parameterizations compound as high-frequency GWs influence atmospheric characteristics including temperature, pressure, and background wind velocities. High-frequency GWs in the MLT region can have large associated momentum fluxes (Fritts et al., 2014). Although these disturbances occur on rapid time scales as compared to those generally utilized by modern GCMs, they occur often enough to have quantifiable impact on underlying atmospheric behavior (Geller et al., 2013; Medvedev and Klaassen, 2000).

1.3.2 Observational constraints

At 80–140 km, in situ satellite observations are unachievable due to proximity to Earth, and high-altitude balloon missions are unfeasible as the MLT remains above their operational ceiling. Remote sensing techniques utilizing OH airglow have been used in recent international space station missions including the Near Infrared Airglow Camera (NIRAC) (Hecht et al., 2023) and the Atmospheric Waves Experiment (AWE) (Zhang et al., 2025). These recent missions have provided insight at mid latitudes, but lack observations in the polar regions.

Ground-based observations from the OH imager at the Poker Flat Research Range (PFRR) provide a complement to satellite measurements with data in a polar region. However, ground-based imagers also face constraints, including contamination from cloud cover, precipitation, the full moon, and daylight. For purposes of data validation discussed in Section 2.4 we filter out times with observable cloud cover. We also note the polar latitude of our imager and restrict observations to September - April to avoid the long Arctic summer daylight.

2 Methods

2.1 Ground-based hydroxyl imaging

Hydroxyl (OH) airglow is a faint, near-uniform, and continuous light emission originating near 87 km in altitude in the MLT (Baker and Jr, 1988). Named the OH Meinel bands for the first person to identify them, it results from the deactivation of vibrationally excited OH molecules, which in turn are produced by exothermic chemical interactions between ozone (${\text{O}}_3$) and atomic hydrogen (H) (Meinel, 1950). This near-infrared emission forms a well-constrained, stable environment that is intimately reliant on small variations in temperature and particle density, including those associated with GWs, instabilities, and auroral activity (Krassovsky and Shagaev, 1977; Wüst et al., 2023). As such, it is an ideal tracer for ground-based imaging (Simkhada et al., 2009).

Since its installation in September 2021, the Near-Earth Space-Sensing Group (NESS) at Arizona State University has managed the OH imager at the PFRR, 65°N 147°W. This imager is equipped with a 185° fisheye lens, allowing for all-sky images covering an area of approximately 1,000,000 ${\text{km}}^2$ in the near-infrared (NIR) between 900–1,700 nm. This enables direct imaging of the OH Meinel band emissions (Berkheimer et al., 2021; Norrell et al., 2024).

This OH imager provides seasonal data for the Gravity Wave Zoo project; an active citizen science initiative on MLT variability associated with GW, instability, and auroral behavior. Images are captured every 10 seconds during nighttime hours between September 1st and April 30th with a 3-s exposure.

2.2 Image processing

Processing begins by un-warping raw images to account for edge distortion attributable to the use of a fisheye lens, and calculating the spatial distance associated with each pixel (Garcia et al., 1997). Images are cropped to a region centered near the zenith (approximately 600 ${\text{km}}^2$), avoiding distorted edge effects. Image processing continues with background subtraction. This is performed using a 70-image running average of emission intensity to emphasize high-frequency atmospheric behavior, highlighting the residual brightness of GWs, instabilities, and short-period auroral arcs. It should be noted that the use of a 70-image running average filters away any activity at a period greater than 15 min (equivalent to the sum of the 70 images’ captures). Examples of processed OH airglow images can be seen in Figure 3.

Once processed, images are grouped into bins of 100 and used to produce 10 s videos (at a frame rate of 10 frames/s). This is done in an effort to improve the approachability of data for amateur volunteers. These videos are then uploaded to a Zooniverse study set, with each set representing a single night of data. Lastly, the study set is released for citizen science classification as part of the Gravity Wave Zoo. To account for inclement weather, nights of data containing clouds or similar foreground contamination are checked by eye and removed before image processing occurs.

2.3 Gravity wave zoo

Launched in fall 2023, the Gravity Wave Zoo leverages citizen science to classify otherwise unmanageably large datasets, utilizing the Zooniverse platform (Simpson et al., 2014). This data spans across several seasons of observation, with the earliest beginning in September 2021, and the most recent concluding in April 2025, representing significant, continuous MLT data. The Gravity Wave Zoo’s consistent and long-scale data enables variability comparisons across years and within seasons at significant temporal resolution.

Provided with tutorial materials and discussion boards moderated by NESS group members, Gravity Wave Zoo participants view videos one-by-one at random, at which time they are prompted with three yes or no questions:

1. Are there any gravity waves present in the video? 2. Are there instabilities in the video? 3. Is there aurora present in the video?

Using a 2/3rd supermajority approach (8/12 responses), final classifications are determined after twelve unique participants view each video. Cases where the supermajority is not reached are categorized as inconclusive, and all results are aggregated and forwarded to NESS for further scientific analysis.

2.4 Statistical validation

For the purposes of validating citizen responses, we adopt a short-term study period, examining complete study sets between 27 December 2023 and 4 February 2024. This period is optimal due to a comparatively high number of consecutive nights with minimal contamination from inclement weather, while still down-scaling the complete dataset so that it is approachable to a small number of researchers. It should be noted that this period is concurrent with a sudden stratospheric warming (SSW) with peak intensity on January 16, so while applicable to the 2023-2024 observation season, further analysis is required before we may consider additional observation seasons. In validating classifications, we temporarily remove ambiguous subjects (video clips where the aforementioned supermajority was not met) from the validation period. We assume that classifications collected for this study period comprise a fair subsample of the September 2023 - April 2024 set of volunteer responses.

The primary goal of statistically analyzing citizen classifications is to ensure volunteers are making a fair effort to accurately classify the videos presented to them (Kosmala et al., 2016; Lukyanenko et al., 2020). In pursuit of this goal, we employ Monte Carlo-based permutation testing to estimate p-values following the approach outlined in Phipson and Smyth (2010). In doing so, we test the following null hypothesis:

There is no correlation between citizen and researcher responses–citizens are likely responding to prompts inaccurately.

We establish a ground truth dataset by independently re-conducting classifications, common in citizen science analysis, e.g., (Swanson et al., 2016). This is then compared to the volunteer classification set to establish an observed match rate $(M_{\mathit{obs}})$. We then randomize citizen classifications relative to our classifications for 100,000 trials $(N=\mathrm{100,000})$, counting all iterations where the resulting randomized match rate $(M_i)$ equals or exceeds the observed match rate. This sum is then divided by the total number of iterations to find our estimate p-value; the fraction of trials in which the random match rate equals or exceeds the observed match rate, using the standard finite-sampling correction. This procedure is repeated three times; once each for GWs, instabilities, and aurora. Formally, the p-value is expressed as $p=\frac{1+{\sum }_{i=1}^N(M_i\ge M_{\text{obs}})}{1+N}$.

We consider the results of this verification in Section 3.2 relative to the commonly accepted threshold for statistically significant results, $p=0.05$. A p-value greater than this threshold indicates little-to-no statistical significance in volunteer classifications, supporting the null-hypothesis. Alternatively, a lesser p-value indicates strong statistical significance–better-than-random citizen engagement.

We further report three standard classification metrics–accuracy, precision, and recall–for each of the three observables to diagnose the prevalence of true positives $(TP)$, true negatives $(TN)$, false positives $(FP)$, and false negatives $(FN)$ within the complete classification dataset, providing insight into volunteer decision-making. Accuracy is defined as the ratio of correct classifications to total classifications, $A=M_{\text{obs}}=\frac{TP+TN}{TP+TN+FP+FN}$. Precision expresses the ratio of correctly identified occurrences to total user-classified occurrences as $P=\frac{TP}{FP+TP}$. Finally, recall measures the ratio of correctly identified occurrences to total actual occurrences as $R=\frac{TP}{FN+TP}$. Consequently, higher precision indicates minimal user-submitted false positives and higher recall indicates minimal user-submitted false negatives.

2.5 Background winds

In drawing analytical connections between citizen classification and external observation, we utilize the Poker Flat Meteor Radar (PFMR) to cross-reference citizen-defined GW occurrences with zonal and meridional background wind components. This enables physical verification of volunteer responses through cross-analysis of co-located observational data.

The PFMR is a SKiYMET installation operational since 2018 and located at the PFRR (65°N 147°W). It transmits at 32.55 MHz and has a pulse repetition frequency of 625 Hz (1.6 m IPP), allowing for detailed observations of burning meteors’ ion trails at five discrete altitudes between 82 and 98 km. Wind velocities are discerned via the Doppler shift of reflected signals as ion trails are displaced by background winds. 3-D locations for each ion trail are then determined via multi-antenna interferometry, and horizontal wind components (zonal and meridional) are reconstructed through applications of a least-squares fit to all Doppler shift measurements within a 3 km bin of the target altitude (Klemm et al., 2019).

3 Results

3.1 Gravity wave zoo participation statistics

At the time of writing, 2,669 unique Gravity Wave Zoo participants have performed 76,612 classifications, comprising 6,330 video clips (633,000 images) of MLT data captured in the high-latitude environment over Alaska. Of those contributors, a dedicated subset engage regularly with the project, returning to classify the most recent data uploads representing a months long commitment. A select few contribute regularly in talk forums, and the most dedicated have accumulated more than 1,000 classifications made during their time as volunteers.

From an engagement standpoint, we consider the Gravity Wave Zoo to be a large success, cultivating an online community and readily improving public participation in a research subfield of growing interest–MLT variability.

3.2 Statistical significance of citizen responses

As highlighted in Figure 1, we find extreme statistical significance in citizen scientist classifications for all three observables during our adopted short-term study period. Specifically, we find p-values of $p=1\times 10^{-5}$ for GWs, $p=1\times 10^{-5}$ for aurora, and $p=1.7\times 10^{-4}$ for instabilities. This represents a significant undercut of the conventional significance threshold, indicating that the probability of random choice explaining citizen scientist classifications is negligible. As such, we reject the null hypothesis defined in Section 2.4. Instead, we find:

Figure 1

Figure 1. Permutation testing results and associated p-values for citizen classifications of GWs (a), aurora (b), and instabilities (c). Observed match rate is denoted by a vertical dashed line, while Gaussian curves represent the distribution of randomized match rates for 100,000 permutations. We note that gaps in distribution curves are the numerical result of a discrete number of possible quotients when calculating p-value. We further note wide gaps between the distribution curve and observed match rate for each observed phenomenon - a visual indicator of strong statistical significance.

There is strong correlation between citizen and researcher responses–citizens are responding to prompts accurately and with fair effort.

Volunteer classifications show varied performance across given metrics. For GWs, accuracy, precision, and recall are 0.82, 0.87, and $0.92$, respectively, increasing confidence in user responses for our primary observable. For aurora, accuracy is 0.58, with very high precision at 0.98, but low recall at 0.24. This indicates very few false positives, but a fair amount of false negatives which may reflect volunteer hesitancy in flagging auroral events. For instabilities, accuracy is 0.58, precision is 0.49, and recall is 0.71. This aligns with expectations given the inherent difficulty of visually classifying these low-frequency, complex phenomena.

3.3 Citizen responses and occurrence rates

3.3.1 Atmospheric gravity waves

Depicted in Figure 2a, according to volunteer classifications, GWs occur in $54.5\%$ of all videos classified in the Gravity Wave Zoo. An additional $36.2\%$ of subjects are reported as inconclusive or uncertain, implying possible disorganized waves with weaker amplitudes and directionality–or potentially no waves at all. Further, no waves are sighted in $9.3\%$ of clips.

Figure 2

Figure 2. Frequency distributions of GWs (a), aurora (b), and instabilities (c) as observed by citizen scientists in the full Gravity Wave Zoo video dataset. Videos are categorized by the proportion of “yes” responses: more than $66.7\%$ are considered confirmed occurrences, fewer than $33.3\%$ are non-occurrences, and $33.3\%$ to $66.7\%$ are classified as inconclusive - cases where no supermajority in responses was reached.

3.3.2 Aurora

With a similar distribution to GWs, aurora (shown in Figure 2b) is sighted in $40.1\%$ of video subjects. Tied with GWs for the lowest portion of inconclusive videos – $36.2\%$ – volunteers likely find clear aurora easiest to classify, possibly due to its bright intensities and unique arcs. Lastly, $23.7\%$ of videos are considered non-occurrences.

3.3.3 Atmospheric wave instabilities

Highlighted in Figure 2c, instabilities are undoubtedly the most difficult phenomenon to classify–expected due to their smaller size and variable intensities. It is noted that classifying instabilities based on images alone is difficult as instabilities can have spatial scales and periods that are close to high-frequency GWs (wavelengths $\approx$ 10 km). Further, the smallest instabilities may elude capture due to restrictive image resolution capabilities. More than half of all subject videos $(52.0\%)$ are recorded as inconclusive, indicating considerable disagreement within the citizen scientist community. Volunteer responses indicate that instabilities are the rarest of the three observables, with only $23.4\%$ of subject clips confirmed as sightings. Similarly, non-occurrences are greatest for instabilities, absent from $24.6\%$ of classified video clips.

4 Discussion

4.1 Challenges associated with citizen science

There are several notable points of error when employing citizen science for image analysis. The primary has foundations in human psychology–how each individual volunteer will think as compared to the next. When faced with a binary decision between reporting an occurrence or non-occurrence, some users will gravitate towards affirming suspicions while others will reject ambiguities in favor of more definite observations. This is exemplified by middle images in Figure 3 where the super majority threshold was either narrowly exceeded or not met. We attempt to mitigate these biases by providing developed, unbiased tutorial materials which include example images, introductory definitions, and room for further education using researcher-moderated talk forums. We further employ said super majority to dictate final determinations, reducing individual perception bias. As highlighted by outermost images in Figure 3, we find reasonable success in this approach, however it must be noted that these biases cannot be erased as they are inherent to human decision making–a contributing factor in the large number of inconclusive subject videos emphasized in Figure 2. We must acknowledge what appears clear to trained researchers may not to informally trained volunteers due to experiential disparities.

Figure 3

Figure 3. An example collection of processed OH-layer images depicting the three target observables: GWs (top row), aurora (middle row), and instabilities (bottom row). The portion of volunteers who responded “yes” to the above images increases from left to right in each row, and is additionally noted below each image. Leftmost images are confirmed non-occurrences, while rightmost images are confirmed occurrences. Instabilities are outlined here for convenience; outlines do not appear in videos presented to citizen scientists.

Reported performance metrics exclude subjects lacking a final classification (cases where a supermajority was not reached). Substantial disagreement among volunteers over the prominence of features in these subjects highlights a key challenge for future work; properly accounting for ambiguous occurrences that are nonetheless scientifically significant. Volunteer classifications may provide crucial information as to how often ambiguous nights of activity are recorded, which contain neither ‘clear’ occurrences, nor ‘quiet’ conditions. While analyzing these nights using either image classification or frequency analysis techniques proves difficult, they remain significant in diagnosing overall GW behavior in the region.

4.2 Associations with background winds

When examining GW occurrences in comparison with background winds recorded by the PFMR, we expect to see characteristic behavior of wave filtering modulated by the vertical structure of the mean flow across altitudes in and around the OH-layer (Fritts and Alexander, 2003). We expect GWs to be most prominent at times where background winds are similar in both magnitude and directionality, as a more cohesive environment is favorable for vertical wave propagation (Fritts and Alexander, 2003). Should magnitude or directionality differ at adjacent altitudes, we may expect weakening GW occurrences due to partial wave dissipation. More extreme discrepancies such as wind reversal or large vertical shears may result in complete suppression of the wave system due to critical filtering (Hines, 1960; Lindzen, 1981).

As exhibited by Figure 4, we observe behavior analogous to expectations for the night of 5–6 January 2024. The night begins with changing directionality over the altitude range in both zonal and meridional directions. Consequently, minimal GW occurrences are classified at the beginning of the observation period between 2:00–4:00 UTC. As the night progresses, background winds in the zonal direction over the altitude range converge in both magnitude and direction, while meridional winds approach zero. This creates favorable conditions for zonal GW propagation. At these times (approximately 5:30–8:00 UTC), citizen scientists sight consistent wave activity. Although Figure 4 represents one particularly strong night of data, we note there exists similar structure in several nights within the adopted 1-month statistical validation period with varying levels of structural clarity.

Figure 4

Figure 4. Zonal and meridional winds for the night of 5–6 January 2024 as recorded by the PFMR. Discrete altitudes observed by the radar system are noted by various dashed lines. Times of citizen-classified GW occurrences are marked by gray overlays. For zonal winds, positive magnitudes denote eastward directionality and negatives denote westward. For meridional winds, positive magnitudes denote northern directionality and negatives denote southern.

4.3 Future research initiatives

4.3.1 Semidiurnal tides

Atmospheric tides are planetary modulations in pressure, temperature, and wind speeds. A proxy for tidal amplitudes may be assumed through sinusoidal fitting or band-pass filtering over PFMR wind measurements. The semidiurnal tide (periods of 12-h) exerts dynamical influences on the MLT in similar processes to GW dispersion. The Gravity Wave Zoo offers a unique opportunity to study GWs in relation to tidal amplitudes, seeking out interactions between the two dynamical aspects on a night-by-night basis to better comprehend long-scale transformations of tidal modality in the MLT.

4.3.2 Sudden stratospheric warmings

SSWs are characterized as sudden increases in polar temperatures and strong disruptions to the circulatory polar vortex over a period of several days. The Gravity Wave Zoo is ideal for use in investigating any interactions between an ongoing SSW and GW propagation, as night-by-night analysis allows for comparisons of wave activity during the onset, peak, and recovery phases of the SSW. Furthermore, multi-year observations allow for comparisons of similar times in different seasons with and without SSWs. This provides a best opportunity for comparison while minimizing annual variables of lesser interest. One such early-stage investigation entails the January 2024 SSW detailed in Zhang et al. (2025), analyzing current hypotheses of reduced GW activity following SSWs.

4.3.3 Direction of wave propagation

GW directionality is imperative to understanding the nuance of wave interactions with the mean flow. Studies of directionality allow for investigation into GW genesis by identifying points of GW origin. While multiple studies exist investigating horizontal wave propagation, they are often restricted by the same temporal scales that limit OH-layer observations (Section 1.3.2). The Gravity Wave Zoo provides consistent nightly data, already formatted with cardinal directions. As such, a new investigation of horizontal wave directionality is a best-use argument for the continuation of citizen science in this context and is in preliminary development. Clear, citizen-classified occurrences of GWs from this work may be used in a tangential project seeking classification on directionality, then undergoing similar analysis procedures as discussed here.

4.4 Conclusions

Evaluating Gravity Wave Zoo statistical significance and its extensive engagement pool is an inherent step in expanding the quality and quantity of MLT variability data for use in future studies. Such long-term research has implications in assessing background winds at various altitudes to highlight semidiurnal atmospheric tidal behavior and influences on GW propagation. Additionally, the dataset can also be used to study impacts of significant phenomenon such as SSWs on small-scale GWs in the MLT region. Through data case studies and validation of citizen science data, we demonstrate both initial findings, as well as the potential uses of this extensive citizen science OH image dataset, summarized as such:

• 2,669 unique participants have engaged with the Gravity Wave Zoo, performing 76,612 classifications on 6,330 video clips (633,000 images) of high-latitude MLT variability data at superior temporal resolution.

• Statistical analysis indicates better-than-random citizen scientist engagement, with p-value estimates of $p=1\times 10^{-5}$, $p=1\times 10^{-5}$, and $p=1.7\times 10^{-4}$ for GWs, aurora, and instabilities, respectively (Figure 1).

• Performance metrics support high confidence in GW classifications, with varied results for aurora and instabilities.

• Overall classifications report GWs in 54.5%, aurora in 40.1%, and instabilities in 23.4% of video clips, with large subject populations classified as inconclusive due to dynamical ambiguities and individual perception bias (Figures 2, 3).

• Several nights (with the most prominent shown in Figure 4) display possible correlation between citizen classifications and favorable wind conditions for the vertical propagation of GWs.

Data availability statement

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

Author contributions

TK: Writing – original draft, Writing – review and editing. KB: Writing – review and editing. JB: Writing – review and editing. JN: Writing – review and editing. SP: Writing – review and editing. KM: Writing – review and editing. P-DP: Writing – review and editing.

Funding

The author(s) declared that financial support was received for this work and/or its publication. This work is supported by US NSF (grant nos. AGS-1944027 and AGS-1936373). The Poker Flat Meteor Radar is supported by US NSF (grant no. AGS-1651464).

Acknowledgements

This publication would not be possible without the contributions of our 2,600+ dedicated citizen scientists who performed classifications for this work. We thank each volunteer for their strong commitment to fostering curiosity and scientific growth. While we cannot acknowledge all dedicated individuals as many prefer anonymity, we would like to recognize one of our most engaged participants who performed hundreds of classifications representing a continuous, months-long commitment: Sallyann Chesson. We thank Denise Thorsen for her continued dedication to collaboration and dutiful management of the Poker Flat Meteor Radar. We further recognize Peter Mason’s contributions in Zooniverse software and data aggregation which continue to prove highly integral to this work. This publication uses data generated via the Zooniverse.org platform, development of which is funded by generous support, including a Global Impact Award from Google, and by a grant from the Alfred P. Sloan Foundation. The Gravity Wave Zoo is found at https://www.zooniverse.org/projects/jberkhei/gravity-wave-zoo. Video classification data used in this work is accessible via Zenodo at https://doi.org/10.5281/zenodo.17888055.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

Alexander, M. J., and Holton, J. R. (1997). A model study of zonal forcing in the equatorial stratosphere by convectively induced gravity waves section. J. Atmos. Sci. doi:10.1175/1520-0469(1997)054<0408:AMSOZF>2.0.CO;2

CrossRef Full Text | Google Scholar

Baker, D. J., and Jr, A. T. S. (1988). Rocket measurements of the altitude distributions of the hydroxyl airglow. Phys. Scr. 37, 611–622. doi:10.1088/0031-8949/37/4/021

CrossRef Full Text | Google Scholar

Baumgarten, G., and Fritts, D. C. (2014). Quantifying kelvin-helmholtz instability dynamics observed in noctilucent clouds: 1. Methods and observations. J. Geophys. Res. 119, 9324–9337. doi:10.1002/2014jd021832

CrossRef Full Text | Google Scholar

Becker, E., Vadas, S. L., Bossert, K., Harvey, V. L., Zülicke, C., and Hoffmann, L. (2022). A high-resolution whole-atmosphere model with resolved gravity waves and specified large-scale dynamics in the troposphere and stratosphere. J. Geophys. Res. Atmos. 127, e2021JD035018. doi:10.1029/2021jd035018

CrossRef Full Text | Google Scholar

Berkheimer, J., Bossert, K., and Pautet, P.-D. (2021). A new OH airglow imager at the poker flat research range for the study of gravity waves, instabilities, and breaking. vol. 2021, SA45A–2186.

Google Scholar

Bossert, K., Fritts, D. C., Pautet, P.-D., Williams, B. P., Taylor, M. J., Kaifler, B., et al. (2015). Momentum flux estimates accompanying multiscale gravity waves over Mount Cook, New Zealand, on 13 July 2014 during the DEEPWAVE campaign. J. Geophys. Res. 120, 9323–9337. doi:10.1002/2015jd023197

CrossRef Full Text | Google Scholar

Dong, W., Fritts, D. C., Liu, A. Z., Lund, T. S., and Liu, H.-L. (2023). Gravity waves emitted from kelvin-helmholtz instabilities. Geophys. Res. Lett. 50, e2022GL102674. doi:10.1029/2022gl102674

CrossRef Full Text | Google Scholar

Fritts, D. C., and Alexander, M. J. (2003). Gravity wave dynamics and effects in the middle atmosphere. Rev. Geophys. 41. doi:10.1029/2001rg000106

CrossRef Full Text | Google Scholar

Fritts, D. C., Pautet, P.-D., Bossert, K., Taylor, M. J., Williams, B. P., Iimura, H., et al. (2014). Quantifying gravity wave momentum fluxes with mesosphere temperature mappers and correlative instrumentation. J. Geophys. Res. 119 (13), 583–603. doi:10.1002/2014jd022150

CrossRef Full Text | Google Scholar

Fritts, D. C., Wang, L., Lund, T. S., Thorpe, S. A., Kjellstrand, C. B., Kaifler, B., et al. (2022). Multi-scale kelvin-helmholtz instability dynamics observed by PMC turbo on 12 July 2018: 2. DNS modeling of KHI dynamics and PMC responses. J. Geophys. Res. 127, e2021JD035834. doi:10.1029/2021jd035834

CrossRef Full Text | Google Scholar

Garcia, F. J., Taylor, M. J., and Kelley, M. C. (1997). Two-dimensional spectral analysis of mesospheric airglow image data. Appl. Opt. 36, 7374–7385. doi:10.1364/AO.36.007374

PubMed Abstract | CrossRef Full Text | Google Scholar

Geller, M. A., Alexander, M. J., Love, P. T., Bacmeister, J., Ern, M., Hertzog, A., et al. (2013). A comparison between gravity wave momentum fluxes in observations and climate models. Boston, MA: American Meteorological Society Journal of Climate.

Google Scholar

Gossard, E., and Munk, W. (1954). On gravity wave in the atmosphere section. Boston, MA: American Meteorological Society Journal of Atmospheric Science.

Google Scholar

Hecht, J. H. (2004). Instability layers and airglow imaging. Rev. Geophys. 42. doi:10.1029/2003rg000131

CrossRef Full Text | Google Scholar

Hecht, J. H., Liu, A. Z., Walterscheid, R. L., and Rudy, R. J. (2005). Maui mesosphere and lower thermosphere (maui MALT) observations of the evolution of kelvin-helmholtz billows formed near 86 km altitude. J. Geophys. Res. 110. doi:10.1029/2003jd003908

CrossRef Full Text | Google Scholar

Hecht, J. H., Gelinas, L. J., Rudy, R. J., and Walterscheid, R. L. (2023). Atmospheric gravity wave and instability observations from the international space station using the near InfraRed airglow camera (NIRAC). J. Geophys. Res. 128, e2023JD039070. doi:10.1029/2023jd039070

CrossRef Full Text | Google Scholar

Hines, C. O. (1960). Internal atmospheric gravity waves at ionospheric heights. Can. J. Phys. 38, 1441–1481. doi:10.1139/p60-150

CrossRef Full Text | Google Scholar

Hunsucker, R. D. (1982). Atmospheric gravity waves generated in the high-latitude ionosphere: a review. Rev. Geophys. 20, 293–315. doi:10.1029/rg020i002p00293

CrossRef Full Text | Google Scholar

Klemm, J., Alspach, J., Li, J., Thorsen, D., Collins, R., Bossert, K., et al. (2019). METR01: new meteor radar at poker flat research range: first observations and analysis.

Google Scholar

Kosmala, M., Wiggins, A., Swanson, A., and Simmons, B. (2016). Assessing data quality in citizen science. Front. Ecol. Environ. 14, 551–560. doi:10.1002/fee.1436

CrossRef Full Text | Google Scholar

Krassovsky, V. I., and Shagaev, M. V. (1977). On the nature of internal gravitational waves observed from hydroxyl emission. Planet. Space Sci. 25, 200–201. doi:10.1016/0032-0633(77)90027-7

CrossRef Full Text | Google Scholar

Lindzen, R. S. (1981). Turbulence and stress owing to gravity wave and tidal breakdown. J. Geophys. Res. Oceans 86, 9707–9714. doi:10.1029/jc086ic10p09707

CrossRef Full Text | Google Scholar

Lukyanenko, R., Wiggins, A., and Rosser, H. K. (2020). Citizen science: an information quality research frontier. Inf. Syst. Front. 22, 961–983. doi:10.1007/s10796-019-09915-z

CrossRef Full Text | Google Scholar

Medvedev, A. S., and Klaassen, G. P. (2000). Parameterization of gravity wave momentum deposition based on nonlinear wave interactions: basic formulation and sensitivity tests. J. Atmos. Solar-Terrestrial Phys. 62, 1015–1033. doi:10.1016/s1364-6826(00)00067-5

CrossRef Full Text | Google Scholar

Meinel, I. A. B. (1950). OH emission bands in the spectrum of the night sky. Astrophysical J. 111, 555. doi:10.1086/145296

CrossRef Full Text | Google Scholar

Norrell, J., Bossert, K., Berkheimer, J., Collins, R. L., Li, J., and Thorsen, D. (2024). Mesospheric atmospheric instabilities above poker flat.

Google Scholar

Oyama, S., and Watkins, B. J. (2012). Generation of atmospheric gravity waves in the polar thermosphere in response to auroral activity. Space Sci. Rev. 168, 463–473. doi:10.1007/s11214-011-9847-z

CrossRef Full Text | Google Scholar

Pardini, C., and Anselmo, L. (2009). Assessment of the consequences of the Fengyun-1C breakup in low Earth orbit. Adv. Space Res. 44, 545–557. doi:10.1016/j.asr.2009.04.014

CrossRef Full Text | Google Scholar

Phipson, B., and Smyth, G. K. (2010). Permutation p-values should never be zero: calculating exact p-values when permutations are randomly drawn. Stat. Appl. Genet. Mol. Biol. 9, Article39. doi:10.2202/1544-6115.1585

PubMed Abstract | CrossRef Full Text | Google Scholar

Plougonven, R., and Zhang, F. (2014). Internal gravity waves from atmospheric jets and fronts. Rev. Geophys. 52, 33–76. doi:10.1002/2012rg000419

CrossRef Full Text | Google Scholar

Simkhada, D. B., Snively, J. B., Taylor, M. J., and Franke, S. J. (2009). Analysis and modeling of ducted and evanescent gravity waves observed in the Hawaiian airglow. Ann. Geophys. 27, 3213–3224. doi:10.5194/angeo-27-3213-2009

CrossRef Full Text | Google Scholar

Simpson, R., Page, K. R., and De Roure, D. (2014). “Zooniverse: observing the world’s largest citizen science platform,” in Proceedings of the 23rd international conference on world wide web (New York, NY, USA: Association for Computing Machinery), 1049–1054. doi:10.1145/2567948.2579215

CrossRef Full Text | Google Scholar

Swanson, A., Kosmala, M., Lintott, C., and Packer, C. (2016). A generalized approach for producing, quantifying, and validating citizen science data from wildlife images. Conserv. Biol. 30, 520–531. doi:10.1111/cobi.12695

PubMed Abstract | CrossRef Full Text | Google Scholar

Turunen, E., Kero, A., Verronen, P. T., Miyoshi, Y., Oyama, S.-I., and Saito, S. (2016). Mesospheric ozone destruction by high-energy electron precipitation associated with pulsating aurora. J. Geophys. Res. Atmos. 121 (11), 852–861. doi:10.1002/2016jd025015

CrossRef Full Text | Google Scholar

Vadas, S. L., Fritts, D. C., and Alexander, M. J. (2003). Mechanism for the generation of secondary waves in wave breaking regions section. J. Atmos. Sci.. doi:10.1175/1520-0469(2003)060<0194:MFTGOS>2.0.CO;2

CrossRef Full Text | Google Scholar

Vadas, S. L., Liu, H.-L., and Lieberman, R. S. (2014). Numerical modeling of the global changes to the thermosphere and ionosphere from the dissipation of gravity waves from deep convection. J. Geophys. Res. 119, 7762–7793. doi:10.1002/2014ja020280

CrossRef Full Text | Google Scholar

Wüst, S., Bittner, M., Espy, P. J., French, W. J. R., and Mulligan, F. J. (2023). Hydroxyl airglow observations for investigating atmospheric dynamics: results and challenges. Atmos. Chem. Phys. 23, 1599–1618. doi:10.5194/acp-23-1599-2023

CrossRef Full Text | Google Scholar

Yue, J., Vadas, S. L., She, C.-Y., Nakamura, T., Reising, S. C., Liu, H.-L., et al. (2009). Concentric gravity waves in the mesosphere generated by deep convective plumes in the lower atmosphere near Fort Collins, Colorado. J. Geophys. Res. 114. doi:10.1029/2008jd011244

CrossRef Full Text | Google Scholar

Zhang, J., Zhao, Y., Pautet, P.-D., Scherliess, L., Taylor, M. J., and Liu, H. (2025). Gravity wave activity during the 2024 sudden stratospheric warmings observed by atmospheric waves experiment (AWE) - Zhang - 2025 - Geophysical research letters. Wiley Online Library.

Google Scholar