Editorial Type:
Article
Article Type:
Research Article

An Accurate and Automated Convective Vortex Detection Method for Long-Duration Infrasound Microbarometer Data

Elizabeth M. Berg aSandia National Laboratories, Albuquerque, New Mexico

Search for other papers by Elizabeth M. Berg in
Current site
Google Scholar
PubMed Close
https://orcid.org/0000-0002-1969-0350
,
Louis J. Urtecho bNASA Jet Propulsion Laboratory, Pasadena, California

Search for other papers by Louis J. Urtecho in
Current site
Google Scholar
PubMed Close
,
Siddharth Krishnamoorthy bNASA Jet Propulsion Laboratory, Pasadena, California

Search for other papers by Siddharth Krishnamoorthy in
Current site
Google Scholar
PubMed Close
,
Elizabeth A. Silber aSandia National Laboratories, Albuquerque, New Mexico

Search for other papers by Elizabeth A. Silber in
Current site
Google Scholar
PubMed Close
,
Andrew Sparks cUniversity of Oregon, Eugene, Oregon

Search for other papers by Andrew Sparks in
Current site
Google Scholar
PubMed Close
, and
Daniel C. Bowman aSandia National Laboratories, Albuquerque, New Mexico

Search for other papers by Daniel C. Bowman in
Current site
Google Scholar
PubMed Close
https://orcid.org/0000-0002-9341-520X
Online Publication:
26 Mar 2024
Print Publication:
01 Mar 2024
DOI:
https://doi.org/10.1175/JTECH-D-23-0037.1
Page(s):
341–354
Open access
View license

Abstract

Heating of the surficial layer of the atmosphere often generates convective vortices, known as “dust devils” when they entrain visible debris. Convective vortices are common on both Earth and Mars, where they affect the climate via dust loading, contribute to wind erosion, impact the efficiency of photovoltaic systems, and potentially result in injury and property damage. However, long-duration terrestrial convective vortex activity records are rare. We have developed a high-precision and high-recall method to extract convective vortex signatures from infrasound microbarometer data streams. The techniques utilizes a wavelet-based detector to capture potential events and then a template matching system to extract the duration of the vortex. Since permanent and temporary infrasound sensors networks are present throughout the globe (many with open data), our method unlocks a vast new convective vortex dataset without requiring the deployment of specialized instrumentation.

Significance Statement

Convective vortices, or “dust devils,” contribute to regional dust loading in Earth’s atmosphere. However, long-duration convective vortex activity records are rare. We came up with a way to autonomously detect the pressure signatures left by convective vortices striking low-frequency sound, or “infrasound,” sensors. Since permanent infrasound stations have been active for decades, our method has the potential to add orders-of-magnitude more events than previously catalogued.

© 2024 American Meteorological Society. This published article is licensed under the terms of the default AMS reuse license. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Elizabeth M. Berg, eliberg@sandia.gov

Abstract

Heating of the surficial layer of the atmosphere often generates convective vortices, known as “dust devils” when they entrain visible debris. Convective vortices are common on both Earth and Mars, where they affect the climate via dust loading, contribute to wind erosion, impact the efficiency of photovoltaic systems, and potentially result in injury and property damage. However, long-duration terrestrial convective vortex activity records are rare. We have developed a high-precision and high-recall method to extract convective vortex signatures from infrasound microbarometer data streams. The techniques utilizes a wavelet-based detector to capture potential events and then a template matching system to extract the duration of the vortex. Since permanent and temporary infrasound sensors networks are present throughout the globe (many with open data), our method unlocks a vast new convective vortex dataset without requiring the deployment of specialized instrumentation.

Significance Statement

Convective vortices, or “dust devils,” contribute to regional dust loading in Earth’s atmosphere. However, long-duration convective vortex activity records are rare. We came up with a way to autonomously detect the pressure signatures left by convective vortices striking low-frequency sound, or “infrasound,” sensors. Since permanent infrasound stations have been active for decades, our method has the potential to add orders-of-magnitude more events than previously catalogued.

© 2024 American Meteorological Society. This published article is licensed under the terms of the default AMS reuse license. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Elizabeth M. Berg, eliberg@sandia.gov
Keywords: Dust or dust storms; Vortices; Acoustic measurements/effects; Infrasound; Pattern detection; Spectral analysis/models/distribution

1. Introduction

Convective vortices, colloquially known as “dust devils” when they sweep up enough material to be visible, are rotating columns of air powered by temperature gradients in the planetary boundary layer. They may persist from minutes to hours and travel several kilometers across the landscape (Lorenz et al. 2015). This phenomenon is common on Earth, particularly in arid regions (Lorenz et al. 2016). While they are significant contributors over regional to global scales on Mars (Klose et al. 2016), they also contribute to atmospheric dust aerosols over local to regional scales on Earth (Klose et al. 2016; Tang et al. 2018). This is a concern for solar power generation, as dust loading can reduce efficiency by 50% in some cases (Adinoyi and Said 2013). Dust devils can mobilize chemical and biological contaminants, creating potential impacts to human health (Mormon and Plumlee 2014). Finally, these can pose a danger to human life and property on rare occasions (Lorenz et al. 2016).

Dust-laden convective vortices are often observed on Mars from both orbital and surface sensors (Lorenz et al. 2016). Since dust loading is a primary driver of the Martian climate (Steakley and Murphy 2016), characterizing convective vortex activity is critical to understanding and predicting the planet’s weather. Furthermore, dust devils influence the Mars’ albedo, change the chemistry of the surface, increase the longevity of surface instrumentation by cleaning dust off of solar panels, and introduce planetary protection concerns by scouring contaminants off equipment and introducing them into the environment (Lorenz and Reiss 2015; Lorenz et al. 2016).

A rich body of research exists on dust devils (e.g., Balme and Greeley 2006; Jackson et al. 2018; Kurgansky 2022), and multiyear convective vortex datasets have been collected using Mars landers and rovers (Charalambous et al. 2021; Newman et al. 2019). However, the most comprehensive terrestrial survey consisted of a handful of sensors in a single area over a few seasons (Jackson and Lorenz 2015). This is a critical gap, because convective vortex activity is difficult to extrapolate to other regions. Lorenz and Jackson (2016) note that dust devil occurrence appears to be a nonstationary process, with local substrate and weather conditions strongly affecting their activity. Indeed, terrestrial studies using pressure sensors report that vortex activity varies by factors of 2 or greater over ranges of as little as a few tens of meters, underscoring the spatiotemporal variability of this phenomenon and making it difficult to extrapolate to larger regions (Lorenz et al. 2015). However, some terrestrial and Martian studies have found that convective vortex counts versus magnitude (defined as maximum pressure drop) follows a power-law distribution, with smaller events happening much more often (Lorenz et al. 2015; Lorenz and Jackson 2015; Steakley and Murphy 2016; Kurgansky 2019).

It just so happens that Earth has a sizeable population of readymade convective vortex detectors in the form of infrasound microbarometers. These pressure sensors are tuned to sound waves below the range of human hearing (<20 Hz, or “infrasound”). Nearly a decade ago, Lorenz and Christie (2015) suggested that the globe-spanning, continuously operating International Monitoring System (IMS) infrasound sensor network could be used for dust devil detection. Other infrasound networks in diverse environments such as deserts (Jones et al. 2015), cities (Bird et al. 2021), and the Arctic (Macpherson et al. 2022) have collected years of data that have not yet been analyzed for convective vortex signatures. Convective vortex signals are a form of background noise in the context of infrasound-focused studies (Lorenz and Christie 2015), but the spatiotemporal characteristics of this noise is not well understood at present.

Here we describe a method to automatically detect and characterize convective vortex signatures. This technique employs wavelet and cross-correlation analyses to capture a wide range of vortex sizes reliably and efficiently. We tested our detector using a large set of synthetics before applying it to 2 months of real-world observational data spanning June through July of 2018. Our sample dataset was extracted from a multiyear network of Hyperion infrasound microbarometers at the Nevada National Security Site (NNSS) in the Mojave Desert (Fig. 1). Although we lack complete visual confirmation of convective vortices on this dataset beyond general comments from field observers, we are able to extract the unique convective vortex signals from the infrasound microbarometer dataset through leverage of previous studies that elucidated the attributes of these signals based on empirical observations (Jackson and Lorenz 2015). However, our technique may be applied to any long-duration record of atmospheric pressure, including the aforementioned years of data collected elsewhere. We highlight a preliminary glimpse of convective vortex activity related to meteorological parameters as a demonstration of the utility of this technique.

Fig. 1.
Fig. 1.

Microbarometer station locations (circles) and nearby weather station (triangle) in the Nevada National Security Site (denoted as a star in the overview map). Microbarometer stations used for background noise analyses are shown in red, and stations analyzed for vortex detections are shown in white.

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0037.1

Convective vortex signatures

The physical structure of convective vortices advected over a pressure sensor result in a time series signature described via a Lorentzian profile (Lorenz and Jackson 2015) as
L(t)=−ΔP1+[2(t−t0)/FWHM]2+B,
where ΔP is the pressure dip, t0 the time of maximum dip, and B the background level. Therefore, they create a pressure drop on stationary microbarometers in their path. Convective vortices can be characterized based on the maximum dip amplitude and the time span over which the pressure is at least half the total drop. This is termed the “full width at half maximum” (FWHM).

Convective vortex impacts result in large, prominent signals on infrasound microbarometers. However, on raw infrasound data, this signal appears as an electrocardiogram-like “heartbeat” shape due to convolution of the convective vortex pressure dip and the instrument response. To obtain the Lorentzian profile of a heartbeat-shaped convective vortex from recorded Hyperion data, we deconvolve the instrument response. For example, a convective vortex signal recorded by station D10M0, included in Fig. 2, contains the characteristic heartbeat shape in the raw data (Fig. 3a). In Fig. 3b, we demonstrate how the signal follows the typical convective vortex Lorentzian-profile pressure dip after removing the instrument response in the frequency domain (Merchant 2015). Additionally, we empirically fit a Lorentzian profile to the convective vortex signal (Fig. 3b), and find that a FWHM of 7 s and maximum pressure dip of 60 Pa best matches the instrument-response-removed data. We then convolve the instrument response to the Lorentzian profile to obtain the heartbeat shape, which is shown over the raw microbarometer data in Fig. 3a.

Fig. 2.
Fig. 2.

Convective vortex pressure signatures over a 1-h period on a subset of microbarometer stations (see Fig. 1) on 4 Jul 2018. Each spike, highlighted in gray, is consistent with the signature of a vortex overrunning a microbarometer. The signal has a characteristic “heartbeat” appearance due to the convolution of a pressure dip with the instrument response. Red scale bar denotes 30 Pa.

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0037.1

Fig. 3.
Fig. 3.

Recorded (blue) and synthetic convective vortex (dashed red line) recorded on microbarometer station D10M0 at 1232 local time 4 Jul 2018, including (a) data prior to instrument response removal (blue) and synthetic generated from convolution of the instrument response with a Lorentzian vortex of 7-s FWHM and 60 Pa dip (dashed red). (b) Data following instrument response removal (blue) and synthetic Lorentzian profile from (a) before convolution with the instrument response.

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0037.1

2. Data characteristics and synthetic dataset generation

In early 2017, a network of Hyperion microbarometers began recording acoustic data over a 2-km-radius circle in a valley bottom at the Nevada National Security Site. All these sensors, while intended for infrasound recording as part of the Dry Alluvium Geology (DAG) campaigns (Larotonda and Townsend 2021) as part of the Source Physics Experiment (SPE) to advance physics-based models linked to underground chemical explosions (Snelson et al. 2013), have proven remarkably effective at capturing vortex activity as demonstrated in Fig. 2. These sensors also included either soaker hoses or high-frequency wind reduction shrouds for wind noise reduction purposes. In this work, we test our convective vortex detection methods on a handful of these sensors (shown in red, see Fig. 1) over the months of June and July in 2018. First, to assess the performance of our detectors, we create a suite of synthetics with typical spectral content and background noise levels within a range found from assessment of the dataset over 2018 (see red stations in Fig. 1).

a. Data spectral and noise content

The physics of wind noise generation on infrasound microbarometers is complex, but power generally scales as frequency raised to a negative exponent (Walker and Hedlin 2010; Brown et al. 2014). We therefore model the observed noise via
A(f)=cfs,
where A is power spectral density, c a scalar constant, and s a negative exponent.

To determine the slope and constant typically observed on data over which we anticipate recording convective vortex signals, we investigate acoustic data recorded at a sampling rate of 500 Hz across 16 stations (see red circles in Fig. 1) from 4 to 9 July 2018 over daylight hours of 1700–2300 UTC, or 1000–1600 local time. We process this trimmed dataset by first removing the instrument response, including the scalar corrections of the datalogger and sensor (Fig. 3a), both provided by the corresponding manufacturer, and deconvolution of the sensor frequency response (Merchant 2015), as shown in Fig. 3b. Then, we filter between 0.001 and 20 Hz (Fig. 3a), compute the power spectral density (Fig. 4d, blue curve), and fit the spectra, following Eq. (2) (Brown et al. 2014), via linear least squares to obtain the scalar constant and exponent between 0.1 and 0.01 Hz (Fig. 4d, magenta line).

Fig. 4.
Fig. 4.

Histogram distributions from a week of data assessed over daylight hours from a variety of stations, including (a) Brownian profile constant, (b) Brownian profile slope, and (c) background noise levels. Note that the mean value is shown as a solid vertical lines and one standard deviation as dashed vertical lines for each distribution. (d) Power spectral density (PSD) of data from station D10M0 over 1700–2300 UTC (blue) and fit Brownian trend (magenta).

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0037.1

Distributions of the values of c and s from all assessed stations and 6-h time period are shown in Figs. 4a and 4b. To determine the background noise level at each of the 16 stations, we first calculate the maximum absolute amplitude from 2-s windows spanning each 6-h time period. Next, we remove values greater than three standard deviations from the initial mean for each distribution to remove high-amplitude signals, which may include convective vortices. Finally, we determine the mean background noise level, the distribution of which we show within Fig. 4c.

b. Synthetic data

Based on the spectral noise content of daylight recordings across the microbarometer subset, we created a suite of 10 000 six-hour-long synthetic barometric time series with 10-Hz sampling rate. We performed Monte Carlo sampling over the distribution of noise parameters in Fig. 4 to determine background noise levels in the synthetics. We used a normal distribution in Log10 space centered at the mean and standard deviation for both the Brownian constant and background noise levels, as shown in Figs. 4a and 4c, respectively. While these distributions are not perfectly Gaussian, we are able to comprehensively sample the model space through our substantial number of synthetics. Since no clear trend is seen in the observed slope, we used a uniform distribution from −1.3 to −0.25. We convolved this background noise signal with the Hyperion instrument response to create “raw” data, which we then scaled in the time domain to a randomly chosen background noise level from the prior distribution.

We then seeded each 6-h synthetic with a random number of vortices. This number was chosen from a Monte Carlo prior normal distribution ranging from 0 to 26, with a mean of 13 and standard deviation of 7. The time of insertion was randomly determined for each vortex within each synthetic, but we ensured that two vortices did not overlap within a synthetic. Previous studies have constrained vortices with FWHM from 10 to 100 s (Jackson and Lorenz 2015; Jackson et al. 2018), and our synthetic data effectively capture the underlying noise conditions for this frequency band (0.1–0.01 Hz). To test the boundaries of our detection methods on synthetic data, we created a series of templates based on varying the FWHM from 1 to 200 s, with pressure dips of 0–100 Pa, which pushes the boundaries of our synthetic data’s capabilities to faithfully replicate real-world noise conditions. After determining the characteristics of each vortex, these parameters were applied into the Lorentzian profile [see Eq. (1)] and convolved with the Hyperion instrument response (Merchant 2015) to obtain the expected heartbeat shape. Specifically, for each convective vortex we first performed a fast Fourier transform (FFT) of the Lorentzian profile, then multiplied the Lorentzian profile with the transfer function in the frequency domain, and finally apply an inverse FFT to convert the signal back into the time domain. After creating each heartbeat-shaped vortex, we added these signals to the 6-h simulated-Hyperion synthetic at the associated chosen time. With this method, we obtained 129 941 total convective vortices randomly inserted across the 10 000 synthetics.

3. Detection scheme

We developed two detection techniques for extracting vortex signatures from the microbarometer dataset. These techniques—a wavelet-based detector and a cross-correlation-based detector—were deployed and characterized individually and in sequence on the synthetic dataset. Then, the techniques were applied to the observational dataset recorded at the NNSS in June and July 2018.

a. Wavelet-based detector

Given their short time duration and unique frequency profiles, convective vortex signatures are prime candidates for detection via spectral analysis. In this section, we discuss the development of a detector that compares the power in the wavelet spectrum of a time series with a background noise model to identify peaks of interest with known statistical confidence intervals.

1) Data preprocessing

Pressure data need to be preprocessed prior to analysis with the wavelet-based detection technique. To do so, any signal anomalies such as isolated unphysical spikes are removed and interpolated over. Then, the instrument response is removed from the time series, which converts the EKG-like signal to a pressure dip (Fig. 3). We experimented with time series with and without instrument response removal and found the performance of the detector to be better when instrument response was removed—doing so avoided the erroneous amplification of noise and attenuation of large convective vortices that occupied lower-frequency bands, improving the detectability of real signals and reducing false detections of noise (Fig. 5). Once instrument response is removed, including scalar corrections of the datalogger and sensor as well as deconvolution to account for sensor frequency response (Merchant 2015), a bandpass filter from 4 mHz to 4 Hz is applied and time series are decimated to 10-Hz sampling rate from their native sampling rates (>100 Hz) to reduce computational time.

Fig. 5.
Fig. 5.

Optimization of the wavelet detector for increased precision and recall. The original time series is plotted in blue, with red sections displaying the “detected” part of the time series. Red contours in the spectrogram indicate the region within which the signal is not produced by background noise with 95% confidence. Columns show time series (left) before and (right) after application of a given step. (a) Removal of instrument response leads to a reduction in the spurious amplification of noise relative to the vortex signal. (b) The wavelet spectrum is smoothed, removing localized pockets of noise-induced high wavelet power. (c) Speckles are rejected when the peak energy is above 0.1 Hz or the “detection” is shorter than 20 s in duration, based on observational data on naturally occurring pressure vortices. (d) Large vortices can influence the signal variance, causing the nondetection of smaller vortices in the same time series. A second pass with wavelet power scaled by outlier-removed variance allows for the detection of smaller vortices. A different time series was used for (d) compared to (a)–(c) to better illustrate the effect of that step.

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0037.1

2) Wavelet transform

We base our wavelet detection scheme on the significance testing in the time–frequency domain discussed in Torrence and Compo (1998, referred to as TC98 hereafter). We employ the complex Morlet wavelet ψ0(η), given by
ψ0(η)=π−1/4eiω0ηe−η2/2,
where ω0 is the nondimensional frequency and is set to 6 to satisfy the admissibility condition for the continuous wavelet transform using Morlet wavelets.
The open-source Python module PyCWT (Krieger et al. 2023) provides a convenient implementation of a correctly scaled Morlet wavelet for our analysis. Significance testing may be conducted by comparing the wavelet power of the signal |Wn(sk)2| scaled by the signal variance σ2 with the background noise power Pk in any frequency bin k. The ratio of these two powers can be compared to a chi-squared distribution to determine the level of confidence (TC98):
|Wn(sk)|2σ2Pk⇒12χ22.
We assume a white-noise background model whereby Pk = 1 for all k. Other noise models such as autocorrelated red noise may be utilized. However, white noise produced satisfactory results in achieving high recall and precision as discussed below.

3) Wavelet detector optimization

The relationship in Eq. (4) is the most basic test to identify peaks of interest in the time series that are not produced by background noise. However, further optimizations may be made to improve the efficiency of the detector by avoiding false positives and false negatives (Fig. 5). These steps are described below:

  1. We smoothed wavelet power across scales to reduce the noise in the wavelet spectrum allowing for fewer false negatives. Following the approach outlined in TC98, wavelet power smoothed along scales is represented by
    W¯n2=δjδtCδ∑j=j1j2|Wn(sj)|2sj,
    where Cδ is a constant, scale-independent reconstruction factor for the complex Morlet wavelet (=0.776), δj is the spacing between the voices of an octave, and j ranges over the scales to be smoothed. We smoothed our wavelet spectrogram across 7-scale windows centered at the scale being evaluated.
    With smoothing along the scale axis, the values of the wavelet coefficients are no longer independent, and Eq. (4) needs to be revised to reflect such scale interdependencies accordingly:
    CδSavgδjδtσ2W¯n2⇒P¯χν2ν,
    where
    Savg=(∑j=j1j21sj)−1,
    P¯=Savg∑j=j1j2Pjsj,
    and the new degree of freedom of the χ2 distribution, ν is given by
    ν=2naSavgSmid1+(naδjδj0)2,
    where na is the number scales used in the averaging, P¯ is the scale-averaged noise power spectrum, and
    Smid=s020.5(j1+j2)δj,
    where the ratio of Savg/Smid corrects for loss of degrees of freedom that arises from dividing the wavelet power spectrum by scale in Eq. (6). Parameter δj0 is chosen to be 0.6 in accordance with TC98, and s0 is the smallest scale chosen to be 0.9549 (corresponding to approximately 1 Hz).

  2. We tune our algorithm to remove all identified peaks in the wavelet spectrum whose duration was less than 20 s or maximum power lay above 0.1 Hz. This is done in heuristic cognizance of the fact that dust devils seen on Earth virtually never occupy this duration or frequency band (Jackson and Lorenz 2015; Lorenz and Lanagan 2014).

  3. Finally, the variance of the signal is affected by any large convective vortices present in the signature. Larger values of variance commensurately scale down the value of the scaled wavelet power in Eq. (6), leading to the missed detection of small convective vortices when large ones are present in the same time series. To avoid this issue, we perform two passes of the wavelet detector on the same time series—first, we run the wavelet detector with the variance computed as is. Once the detector has been run on a given time series, identified detection windows are removed from the computation of the variance and the confidence testing procedure is repeated a second time, with the wavelet power scaled by the new variance. The wavelet spectrum itself does not need to be recalculated. Thus, the additional cost of running the confidence testing step twice is minimal.

b. Correlation-based detector

While the distinct spectral content of convective vortices is highly conducive to a wavelet detector, we also developed and analyzed a cross-correlation-based detector employing synthetic vortex templates across microbarometer data. The application of a correlation detector enables the quantification of the similarity of a waveform shape to a given template, allowing us to leverage the distinct heartbeat shape of the convective vortices as recorded in the raw waveform. To retain the heartbeat shape, we only apply the scalar datalogger bitweight and sensor volts-to-pascals corrections for the collected (nonsynthetic) data processed by the correlation-based detector. Template-matching cross correlation has been widely used in seismology to detect and characterize signals of interest from seismic data using template waveforms, including to detect and characterize low-magnitude earthquake and mining events (e.g.,. Downey et al. 2023; Gibbons and Ringdal 2006; Slinkard et al. 2013; Sundermier et al. 2022). This technique has also been applied to detect vortex encounters through a steady-state modified Lorentzian profile directly (Jackson et al. 2021) without incorporation of the heartbeat shape explored by this study. The correlation detector also has an increased ability to detect vortices adjacent in time, which may otherwise be difficult to separate with the wavelet detector alone. In addition, the wavelet detector identifies all times when the signal has not been generated by background noise with a given confidence level. Not all of these may be related to a vortex—the correlation detector can extract detections specifically associated with vortices.

To obtain our heartbeat-vortex template, we form 15 Lorentzian vortices equally spaced from 1- to 200-s FWHM all with a dip of 100 Pa (see Fig. 6a) and then convolve the Hyperion instrument response to each profile (Fig. 6b). In Fig. 6b a clear decrease in signal amplitude is observed, which is introduced by the instrument response. As the correlation technique is only sensitive to the phase information and not the amplitude of the signal, the decrease in signal strength within the templates does not affect detection outcomes (Fig. 6a). However, the loss of signal strength after instrument response convolution indicates that the Hyperion sensors are less likely to record larger FWHM vortices. We note that our method will not be able to successfully extract a signal whose pressure dip is significantly below the noise threshold of the instrument. Importantly, detectable dust-laden vortices are associated with 10 to 100 Pa pressure dips (Lorenz and Jackson 2015; Jackson et al. 2018), which are well within the sensitivity range of the Hyperion instrument. For each template and data segment, we calculate the correlation coefficient (CC) following Gibbons and Ringdal (2006),
C(t)=template(t)⋅datastream(t)template(t)⋅template(t)datastream(t)⋅datastream(t),
where the dot operator denotes cross correlation. As cross correlation in the time domain is simply the complex conjugate of the template multiplied by the data stream in the frequency domain, we perform all such operations in the frequency domain.
Fig. 6.
Fig. 6.

Record sections for a variety of convective vortex theoretical (a) Lorentzian and (b) heartbeat profiles arranged according to FWHM values, from low (bottom profiles) to high (top profiles), and shaded according to maximum absolute signal strength. These are the templates used in the correlation detector.

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0037.1

Our correlation detection method is described in a step-by-step manner below.

  1. For each 6-h segment of data, we remove the signal mean and linear trend, implement scalar datalogger and sensor corrections, apply a 1% taper to the data, and bandpass filter to retain signals within 0.5–1000 s.

  2. For each template that will be matched, we cut the 6-h segment of data into snippets. Each snippet is twice the template length, and snippets are arranged in a way that neighboring snippets overlap by half a template length.

  3. To best extract potential signals of interest, we filter each snippet to focus on the template FWHM of interest, with the passband set from the FWHM to 1000 s.

  4. For each data snippet,

    1. We perform the mean removal, detrending, and tapering steps again to remove any short-term trends introduced by the long-term trend removal on the 6-h segment, which further stabilizes the FFT.

    2. We compute the correlation coefficient across the snippet [Eq. (11)] with each one of the templates. We improve the efficiency of this process first by computing cross correlation in the frequency domain and then performing an inverse FFT to obtain C(t). Thus, each data segment generates as many C(t) curves as the number of templates.

    3. We time shift each C(t) curve by Δt = tpeak − tstart, the time difference between the start of the template and the time of absolute peak pressure signal for that template. Our eventual aim is to compute the maximum value of C(t) across the different templates. Since each template is of a different length (see Fig. 6), shifting the C(t) curves by Δt ensures that the comparison across templates is consistent.

    4. We compute the maximum of C(t) across templates and retain only the C(t) curve that contains the maximum value and record the template FWHM.

  5. Finally, for the selected C(t) curve, we find the maximum peak value in 50-s running windows. In spans of time where the value of C(t) is above a very conservative threshold of 0.08, we record a “preliminary” detection for further investigation. Importantly, this preliminary value is not the final threshold, but this low threshold enables investigation of performance over a wide range of threshold values.

  6. For each preliminary detection, we calculate the signal-to-noise ratio where the noise window is created from 500 s of data immediately before and after the maximum template match. We provide an example of the correlation detection, including true, false, and missed detections as a function of time in Fig. S1 in the online supplemental material. From empirical evaluation, we determine final detections from the preliminary detections with C(t) and signal-to-noise ratio (SNR) values over thresholds of 0.36 and 4, respectively, as applied in Fig. S1. This last step significantly shortens the periods of time that are marked as detections in the final detection reported, as compared to the preliminary detections.

c. Wavelet + correlation detection

To gain the benefits of both the wavelet and correlation detectors, we run them in sequence. First, we use the wavelet-based detector to identify windows with peak amplitudes. Then we run our correlation detector over these identified windows, expanding the window size to 400 s to accommodate the largest template width (see Fig. 6). The main advantages of this combination is the ability to extract vortex signatures that are near each other in time and ensure the detections match those expected for a vortex. Another advantage of sequencing the detectors is that the wavelet detector significantly reduces the length of time over which a correlation must be run (e.g., Fig. 5), thus saving computational time. In Fig. 7, we show the results of both the wavelet and correlation detectors (Figs. 7a,b, respectively), as well as a sequential application of both detectors with the wavelet detector identifying the time snippets to run the correlation detector (Fig. 7c). As shown, by combining both detectors, we strongly decrease false detections while increasing true detections. Both detectors are also parallelized and can be run on supercomputing clusters. We apply this technique to all synthetics to fully quantify the performance of each detector alone and in combination before applying the optimal parameters to the recorded data from the NNSS.

Fig. 7.
Fig. 7.

Example 6-h synthetic seeded with heartbeat vortices and detections showing results from the (a) wavelet, (b) correlation, and (c) wavelet detector output as input for the correlation detector. The true detections are green windows for the wavelet detector and as overlaid templates and filled circles for correlation coefficient. False detections are red windows (wavelet detector) with a red × at the peak amplitude of the detection (wavelet and correlation detectors). Missed detections are shown as gray triangles.

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0037.1

4. Results

a. Synthetics

We apply each detector to all 10 000 six-hour synthetics in order to assess the trade-off of precision and recall for the final SNR and CC thresholds that determine the final set of potential detections from all preliminary detections. Specifically, recall is related to the number of true compared to missed, or false negative, detections,
Recall=True positivesTrue positives+False negatives,
while precision is related to number of true compared to false positive detections,
Precision=True positivesTrue positives+False positives.
We calculated precision and recall for each detector over the 10 000 synthetics (see Table 1) containing a total of 129 941 inserted synthetic convective vortex signals. We found that the wavelet detector has higher recall (lower rate of missed detections), but the correlation detector has higher precision (lower rate of false detections). By using the output of the wavelet detector as input to the correlation detector, we are able to obtain high recall and the highest precision. We use cross correlation and SNR ratio values of 0.36 and 4, respectively, for both the correlation detector alone and following the wavelet detector. Although these values are relatively low, raising them resulted in more missed detections without a significant decrease in false positives (see Fig. S2). As our wavelet + correlation detector performs exceedingly well, we are confident in our choices for the correlation detector’s parameters.
Table 1.

Summary of performance of detectors over synthetic data.
Table 1.

To observe how performance varies across the seeded convective vortex parameters (see Fig. 8), we first consider recall (Figs. 8a,b,d) for each detector over the range of synthetic FWHM and dip values of inserted convective vortices. While the wavelet detector outperforms the cross-correlation and combined detector, we note that this detector struggles with lower FWHM vortices when these occur near each other in time. When we combine the wavelet detector with the correlation detector, this effect is diminished. The wavelet detector is not able to detect vortex signals with duration of <20 s due to tuning of the wavelet detector to focus on longer-period signals, as dust devils on Earth typically are not observed below this value (Jackson and Lorenz 2015; Lorenz and Lanagan 2014). This causes a decrease in precision for FWHM < 10 s, regardless of dip value. Long-duration signals, often mistaken for vortex encounters (Jackson and Lorenz 2015), and failing to detect signals when near each other in time contribute to lower overall precision for the wavelet detector, shown in Table 1. Similarly, we evaluate precision performance for the correlation (Fig. 8c) and wavelet + correlation (Fig. 8e) detectors according to the detected FWHM. As the wavelet detector does not estimate FWHM or dip values, we do not evaluate performance of false detections from the wavelet detector within Fig. 8. We observed that the correlation detector has a large number of false positives with decreasing cross correlation and increasing FWHM. However, by combining the wavelet and correlation detectors, we reduce the number of false positives and obtain high precision with the relatively low CC and SNR thresholds.

Fig. 8.
Fig. 8.

Performance metrics of detected and inserted convective vortices across the 10 000 six-hour synthetics, including (a) achievable recall of the wavelet detector, (b) achievable recall of the correlation detector, (c) precision of the correlation detector, (d) achievable recall of the combined wavelet and correlation detector, and (e) precision of the wavelet and correlation detector. Note that achievable recall is shown as a function of the inserted synthetic convective vortex FWHM and dip parameters, but precision is shown as a function of the detected FWHM parameter and correlation coefficient (CC).

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0037.1

b. Infrasound microbarometer data

We then applied the wavelet + correlation detector to the dataset collected at the NNSS. We analyze a subset of stations over the months of June and July in 2018. We follow identical preprocessing for each detector as previously described. Prior to applying our detector pipeline, we removed isolated, nonphysical data spikes using a short-term-average to long-term-average detector of 0.1 and 10 s applied to all data with peak values over 45 Pa (prior to instrument response removal).

In Fig. 9a, we show an example result from a single station on 23 June 2018, overlaying the detection templates on the day of recorded pressure data. We observe clear detections between 0900 and 1200 LT, also clearly visible to the human eye with the expected heartbeat-like shape. Both of the detected convective vortices, with peak values above 16 Pa, contain correlation coefficients above 0.95 for the 15 FWHM template. To further showcase our detectors and the data, we also include an example of data that contain an undesirable spike (see Fig. S3a) and the results of our preprocessing and detectors (Fig. S3b).

Fig. 9.
Fig. 9.

Day plot of time series pressure data from station D5M02 on 23 Jun 2018, with detection templates overlaid in gray.

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0037.1

The application of our detector to the 2-month-long dataset resulted in 1785 convective vortices. We analyzed the properties of all detections across the June through July time period, including the local time at which detections occurred, the FWHM of the detections, as well as atmospheric temperature and humidity (Fig. 10). To obtain temperature and humidity, we interpolate 15-min interval data from a nearby weather station (white triangle in Fig. 1).

Fig. 10.
Fig. 10.

Summary of detection statistics over all analyzed stations in June and July, including (a) local time, (b) temperature, (c) humidity, and (d) FWHM of the detections, which contain increased detections with FWHM > 10 s due to current wavelet detector tuning.

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0037.1

5. Conclusions

Convective vortices have key impacts on climate, boundary layer dynamics, dust loading, and even human activity on both Earth and Mars. Rather surprisingly, the convective vortex observation record is actually more complete on Mars than our own planet, which highlights the need for long-term terrestrial studies. We followed up on the suggestion by Lorenz and Christie (2015) to investigate whether infrasound microbarometer data contain extractable evidence of convective vortex activity. We found that a combination of wavelet and cross-correlation methods could detect and characterize simulated convective vortex signatures. We validated our detector against real vortex signatures collected in the Nevada desert.

The techniques described here can unlock convective vortex signatures in years of data collected by hundreds of infrasound microbarometers across the globe. Variations across diverse environments, such as those not usually thought to be susceptible to this activity (e.g., forests, tundra), can now be assessed. This technique holds the promise to vastly increase the repository of convective vortices available to study and associated with atmospheric parameters.

We also look forward to validations of modeled and observed convective vortex activity in cities like that reported in Fujiwara et al. (2011). We are presently engaged in scanning the entire DAG dataset, of which the example data presented here are only a small fraction. We expect to recover on the order of 104 to perhaps even 105 vortices during this effort and fully assess the time, amplitude, and duration of detected vortices in addition to impact of climatological and topographical conditions across the network.

Acknowledgments.

The authors would like to acknowledge financial support from the NASA Solar System Workings (SSW) program. The authors also thank Sarah Albert for her valuable insight and constructive feedback on our study. Part of this research was performed at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration (80NM0018D0004). The High Performance Computing resources used in this investigation were provided by funding from the JPL Information and Technology Solutions Directorate. This article has been authored by an employee of National Technology and Engineering Solutions of Sandia, LLC under Contract DE-NA0003525 with the U.S. Department of Energy (DOE). The employee owns all right, title, and interest in and to the article and is solely responsible for its contents. The U.S. Government retains and the publisher, by accepting the article for publication, acknowledges that the U.S. Government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this article or allow others to do so, for U.S. Government purposes.

Data availability statement.

All microbarometer pressure and weather data analyzed in this article are available for review and will be made openly available on Drayd under DOI: https://doi.org/10.5061/dryad.0cfxpnw7f. Waveform data were processed using external Python libraries NumPy, SciPy, and Obspy (Beyreuther et al. 2010; Harris et al. 2020; Krischer et al. 2015; Virtanen et al. 2020). Additional analysis was performed following methodology presented in this paper. The DOE will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan: https://www.energy.gov/downloads/doe-public-access-plan. This paper describes objective technical results and analysis. Any subjective views or opinions that might be expressed in the paper do not necessarily represent the views of the U.S. Department of Energy or the U.S. Government.

REFERENCES

  • Adinoyi, M. J., and S. A. M. Said, 2013: Effect of dust accumulation on the power outputs of solar photovoltaic modules. Renewable Energy, 60, 633–636, https://doi.org/10.1016/j.renene.2013.06.014.

    • Search Google Scholar
    • Export Citation

  • Balme, M., and R. Greeley, 2006: Dust devils on Earth and Mars. Rev. Geophys., 44, RG3003, https://doi.org/10.1029/2005RG000188.

  • Beyreuther, M., R. Barsch, L. Krischer, T. Megies, Y. Behr, and J. Wassermann, 2010: ObsPy: A Python toolbox for seismology. Seismol. Res. Lett., 81, 530–533, https://doi.org/10.1785/gssrl.81.3.530.

    • Search Google Scholar
    • Export Citation

  • Bird, E., D. C. Bowman, D. R. Seastrand, M. A. Wright, J. M. Lees, and F. K. Dannemann Dugick, 2021: Monitoring changes in human activity during the COVID-19 shutdown in Las Vegas using infrasound microbarometers. J. Acoust. Soc. Amer., 149, 1796–1802, https://doi.org/10.1121/10.0003777.

    • Search Google Scholar
    • Export Citation

  • Brown, D., L. Ceranna, M. Prior, P. Mialle, and R. J. Le Bras, 2014: The IDC seismic, hydroacoustic and infrasound global low and high noise models. Pure Appl. Geophys., 171, 361–375, https://doi.org/10.1007/s00024-012-0573-6.

    • Search Google Scholar
    • Export Citation

  • Charalambous, C., and Coauthors, 2021: Vortex-dominated Aeolian activity at InSight’s landing site, Part 1: Multi-instrument observations, analysis, and implications. J. Geophys. Res. Planets, 126, e2020JE006757, https://doi.org/10.1029/2020JE006757.

    • Search Google Scholar
    • Export Citation

  • Downey, N., S. Albert, and R. Tibi, 2023: The Redmond Salt Mine Monitoring Experiment: Observations of infrasound resonance. Bull. Seismol. Soc. Amer., 113, 1664–1681, https://doi.org/10.1785/0120220114.

    • Search Google Scholar
    • Export Citation

  • Fujiwara, C., K. Yamashita, M. Nakanishi, and Y. Fujiyoshi, 2011: Dust devil-like vortices in an urban area detected by a 3D scanning Doppler lidar. J. Appl. Meteor. Climatol., 50, 534–547, https://doi.org/10.1175/2010JAMC2481.1.

    • Search Google Scholar
    • Export Citation

  • Gibbons, S. J., and F. Ringdal, 2006: The detection of low magnitude seismic events using array-based waveform correlation. Geophys. J. Int., 165, 149–166, https://doi.org/10.1111/j.1365-246X.2006.02865.x.

    • Search Google Scholar
    • Export Citation

  • Harris, C. R., and Coauthors, 2020: Array programming with NumPy. Nature, 585, 357–362, https://doi.org/10.1038/s41586-020-2649-2.

  • Jackson, B., and R. Lorenz, 2015: A multiyear dust devil vortex survey using an automated search of pressure time series. J. Geophys. Res. Planets, 120, 401–412, https://doi.org/10.1002/2014JE004712.

    • Search Google Scholar
    • Export Citation

  • Jackson, B., R. Lorenz, and K. Davis, 2018: A framework for relating the structures and recovery statistics in pressure time-series surveys for dust devils. Icarus, 299, 166–174, https://doi.org/10.1016/j.icarus.2017.07.027.

    • Search Google Scholar
    • Export Citation

  • Jackson, B., J. Crevier, M. Szurgot, R. Battin, C. Perrin, and S. Rodriguez, 2021: Inferring vortex and dust devil statistics from InSight. Planet. Sci. J., 2, 206, https://doi.org/10.3847/PSJ/ac260d.

    • Search Google Scholar
    • Export Citation

  • Jones, K. R., R. W. Whitaker, and S. J. Arrowsmith, 2015: Modeling infrasound signal generation from two underground explosions at the source physics experiment using the Rayleigh integral. Geophys. J. Int., 200, 779–790, https://doi.org/10.1093/gji/ggu433.

    • Search Google Scholar
    • Export Citation

  • Klose, M., and Coauthors, 2016: Dust devil sediment transport: From lab to field to global impact. Space Sci. Rev., 203, 377–426, https://doi.org/10.1007/s11214-016-0261-4.

    • Search Google Scholar
    • Export Citation

  • Krieger, S., and Coauthors, 2023: PyCWT: Spectral analysis using wavelets in Python. PyCWT, https://pycwt.readthedocs.io/en/latest/.

  • Krischer, L., T. Megies, R. Barsch, M. Beyreuther, T. Lecocq, C. Caudron, and J. Wassermann, 2015: ObsPy: A bridge for seismology into the scientific Python ecosystem. Comput. Sci. Discovery, 8, 014003, https://doi.org/10.1088/1749-4699/8/1/014003.

    • Search Google Scholar
    • Export Citation

  • Kurgansky, M. V., 2019: On the statistical distribution of pressure drops in convective vortices: Applications to Martian dust devils. Icarus, 317, 209–214, https://doi.org/10.1016/j.icarus.2018.08.004.

    • Search Google Scholar
    • Export Citation

  • Kurgansky, M. V., 2022: Statistical distribution of atmospheric dust devils on Earth and Mars. Bound.-Layer Meteor., 184, 381–400, https://doi.org/10.1007/s10546-022-00713-w.

    • Search Google Scholar
    • Export Citation

  • Larotonda, J. M., and M. J. Townsend, 2021: Data release report for the Source Physics Experiment phase II: Dry Alluvium Geology experiments (DAG-1 through DAG-4), Nevada National Security Site. NNSS Rep. DOE/NV/03624-1220, 84 pp., https://www.osti.gov/servlets/purl/1825534.

  • Lorenz, R. D., and P. D. Lanagan, 2014: A barometric survey of dust-devil vortices on a desert playa. Bound.-Layer Meteor., 153, 555–568, https://doi.org/10.1007/s10546-014-9954-y.

    • Search Google Scholar
    • Export Citation

  • Lorenz, R. D., and D. Christie, 2015: Dust devil signatures in infrasound records of the International Monitoring System. Geophys. Res. Lett., 42, 2009–2014, https://doi.org/10.1002/2015GL063237.

    • Search Google Scholar
    • Export Citation

  • Lorenz, R. D., and B. K. Jackson, 2015: Dust devils and dustless vortices on a desert playa observed with surface pressure and solar flux logging. GeoResJ, 5, 1–11, https://doi.org/10.1016/j.grj.2014.11.002.

    • Search Google Scholar
    • Export Citation

  • Lorenz, R. D., and D. Reiss, 2015: Solar panel clearing events, dust devil tracks, and in-situ vortex detections on Mars. Icarus, 248, 162–164, https://doi.org/10.1016/j.icarus.2014.10.034.

    • Search Google Scholar
    • Export Citation

  • Lorenz, R. D., and B. K. Jackson, 2016: Dust devil populations and statistics. Space Sci. Rev., 203, 277–297, https://doi.org/10.1007/s11214-016-0277-9.

    • Search Google Scholar
    • Export Citation

  • Lorenz, R. D., L. D. Neakrase, and J. D. Anderson, 2015: In-situ measurement of dust devil activity at Ja Jornada Experimental Range, New Mexico, USA. Aeolian Res., 19, 183–194, https://doi.org/10.1016/j.aeolia.2015.01.012.

    • Search Google Scholar
    • Export Citation

  • Lorenz, R. D., and Coauthors, 2016: History and applications of dust devil studies. Space Sci. Rev., 203, 5–37, https://doi.org/10.1007/s11214-016-0239-2.

    • Search Google Scholar
    • Export Citation

  • Macpherson, K. A., J. R. Coffey, A. J. Witsil, D. Fee, S. Holtkamp, S. Dalton, H. McFarlin, and M. West, 2022: Ambient infrasound noise, station performance, and their relation to land cover across Alaska. Seismol. Res. Lett., 93, 2239–2258, https://doi.org/10.1785/0220210365.

    • Search Google Scholar
    • Export Citation

  • Merchant, B. J., 2015: Hyperion 5113/GP infrasound sensor evaluation. Sandia Rep. SAND2015-7075, 43 pp., https://doi.org/10.2172/1213302.

  • Mormon, S. A., and G. S. Plumlee, 2014: Dust and human health. Mineral Dust: A Key Player in the Earth System, P. Knippertz and J. B. Stuut, Eds., Springer, 385–410, https://doi.org/10.1007/978-94-017-8978-3_15.

  • Newman, C. E., H. Kahanp, M. I. Richardson, G. M. Martínez, A. Vicente-Retortillo, and M. T. Lemmon, 2019: MarsWRF convective vortex and dust devil predictions for Gale Crater over 3 Mars years and comparison with MSL-REMS observation. J. Geophys. Res. Planets, 124, 3442–3468, https://doi.org/10.1029/2019JE006082.

    • Search Google Scholar
    • Export Citation

  • Slinkard, M. E., D. B. Carr, and C. J. Young, 2013: Applying waveform correlation to three aftershock sequences. Bull. Seismol. Soc. Amer., 103, 675–693, https://doi.org/10.1785/0120120058.

    • Search Google Scholar
    • Export Citation

  • Snelson, C. M., R. E. Abbott, S. T. Broome, R. J. Mellors, H. J. Patton, A. J. Sussman, and W. R. Walter, 2013: Chemical explosion experiments to improve nuclear test monitoring. Eos, Trans. Amer. Geophys. Union, 94, 237–239, https://doi.org/10.1002/2013EO270002.

    • Search Google Scholar
    • Export Citation

  • Steakley, K., and J. Murphy, 2016: A year of convective vortex activity at Gale Crater. Icarus, 278, 180–193, https://doi.org/10.1016/j.icarus.2016.06.010.

    • Search Google Scholar
    • Export Citation

  • Sundermier, A., R. Tibi, R. A. Brogan, and C. J. Young, 2022: Applying waveform correlation to reduce seismic analyst workload due to repeating mining blasts. Bull. Seismol. Soc. Amer., 112, 171–190, https://doi.org/10.1785/0120210124.

    • Search Google Scholar
    • Export Citation

  • Tang, Y., Y. Han, and Z. Liu, 2018: Temporal and spatial characteristics of dust devils and their contribution to the aerosol budget in East Asia—An analysis using a new parameterization scheme for dust devils. Atmos. Environ., 182, 225–233, https://doi.org/10.1016/j.atmosenv.2018.03.050.

    • Search Google Scholar
    • Export Citation

  • Torrence, C., and G. P. Compo, 1998: A practical guide to wavelet analysis. Bull. Amer. Meteor. Soc., 79, 61–78, https://doi.org/10.1175/1520-0477(1998)079<0061:APGTWA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation

  • Virtanen, P., and Coauthors, 2020: SciPy 1.0: Fundamental algorithms for scientific computing in Python. Nat. Methods, 17, 261–272, https://doi.org/10.1038/s41592-019-0686-2.

    • Search Google Scholar
    • Export Citation

  • Walker, K. T., and M. A. Hedlin, 2010: A review of wind-noise reduction technologies. Infrasound Monitoring for Atmospheric Studies, Springer, 141–182.

Supplementary Materials

Save
Cover Journal of Atmospheric and Oceanic Technology
Journal of Atmospheric and Oceanic Technology

  • Adinoyi, M. J., and S. A. M. Said, 2013: Effect of dust accumulation on the power outputs of solar photovoltaic modules. Renewable Energy, 60, 633–636, https://doi.org/10.1016/j.renene.2013.06.014.

    • Search Google Scholar
    • Export Citation

  • Balme, M., and R. Greeley, 2006: Dust devils on Earth and Mars. Rev. Geophys., 44, RG3003, https://doi.org/10.1029/2005RG000188.

  • Beyreuther, M., R. Barsch, L. Krischer, T. Megies, Y. Behr, and J. Wassermann, 2010: ObsPy: A Python toolbox for seismology. Seismol. Res. Lett., 81, 530–533, https://doi.org/10.1785/gssrl.81.3.530.

    • Search Google Scholar
    • Export Citation

  • Bird, E., D. C. Bowman, D. R. Seastrand, M. A. Wright, J. M. Lees, and F. K. Dannemann Dugick, 2021: Monitoring changes in human activity during the COVID-19 shutdown in Las Vegas using infrasound microbarometers. J. Acoust. Soc. Amer., 149, 1796–1802, https://doi.org/10.1121/10.0003777.

    • Search Google Scholar
    • Export Citation

  • Brown, D., L. Ceranna, M. Prior, P. Mialle, and R. J. Le Bras, 2014: The IDC seismic, hydroacoustic and infrasound global low and high noise models. Pure Appl. Geophys., 171, 361–375, https://doi.org/10.1007/s00024-012-0573-6.

    • Search Google Scholar
    • Export Citation

  • Charalambous, C., and Coauthors, 2021: Vortex-dominated Aeolian activity at InSight’s landing site, Part 1: Multi-instrument observations, analysis, and implications. J. Geophys. Res. Planets, 126, e2020JE006757, https://doi.org/10.1029/2020JE006757.

    • Search Google Scholar
    • Export Citation

  • Downey, N., S. Albert, and R. Tibi, 2023: The Redmond Salt Mine Monitoring Experiment: Observations of infrasound resonance. Bull. Seismol. Soc. Amer., 113, 1664–1681, https://doi.org/10.1785/0120220114.

    • Search Google Scholar
    • Export Citation

  • Fujiwara, C., K. Yamashita, M. Nakanishi, and Y. Fujiyoshi, 2011: Dust devil-like vortices in an urban area detected by a 3D scanning Doppler lidar. J. Appl. Meteor. Climatol., 50, 534–547, https://doi.org/10.1175/2010JAMC2481.1.

    • Search Google Scholar
    • Export Citation

  • Gibbons, S. J., and F. Ringdal, 2006: The detection of low magnitude seismic events using array-based waveform correlation. Geophys. J. Int., 165, 149–166, https://doi.org/10.1111/j.1365-246X.2006.02865.x.

    • Search Google Scholar
    • Export Citation

  • Harris, C. R., and Coauthors, 2020: Array programming with NumPy. Nature, 585, 357–362, https://doi.org/10.1038/s41586-020-2649-2.

  • Jackson, B., and R. Lorenz, 2015: A multiyear dust devil vortex survey using an automated search of pressure time series. J. Geophys. Res. Planets, 120, 401–412, https://doi.org/10.1002/2014JE004712.

    • Search Google Scholar
    • Export Citation

  • Jackson, B., R. Lorenz, and K. Davis, 2018: A framework for relating the structures and recovery statistics in pressure time-series surveys for dust devils. Icarus, 299, 166–174, https://doi.org/10.1016/j.icarus.2017.07.027.

    • Search Google Scholar
    • Export Citation

  • Jackson, B., J. Crevier, M. Szurgot, R. Battin, C. Perrin, and S. Rodriguez, 2021: Inferring vortex and dust devil statistics from InSight. Planet. Sci. J., 2, 206, https://doi.org/10.3847/PSJ/ac260d.

    • Search Google Scholar
    • Export Citation

  • Jones, K. R., R. W. Whitaker, and S. J. Arrowsmith, 2015: Modeling infrasound signal generation from two underground explosions at the source physics experiment using the Rayleigh integral. Geophys. J. Int., 200, 779–790, https://doi.org/10.1093/gji/ggu433.

    • Search Google Scholar
    • Export Citation

  • Klose, M., and Coauthors, 2016: Dust devil sediment transport: From lab to field to global impact. Space Sci. Rev., 203, 377–426, https://doi.org/10.1007/s11214-016-0261-4.

    • Search Google Scholar
    • Export Citation

  • Krieger, S., and Coauthors, 2023: PyCWT: Spectral analysis using wavelets in Python. PyCWT, https://pycwt.readthedocs.io/en/latest/.

  • Krischer, L., T. Megies, R. Barsch, M. Beyreuther, T. Lecocq, C. Caudron, and J. Wassermann, 2015: ObsPy: A bridge for seismology into the scientific Python ecosystem. Comput. Sci. Discovery, 8, 014003, https://doi.org/10.1088/1749-4699/8/1/014003.

    • Search Google Scholar
    • Export Citation

  • Kurgansky, M. V., 2019: On the statistical distribution of pressure drops in convective vortices: Applications to Martian dust devils. Icarus, 317, 209–214, https://doi.org/10.1016/j.icarus.2018.08.004.

    • Search Google Scholar
    • Export Citation

  • Kurgansky, M. V., 2022: Statistical distribution of atmospheric dust devils on Earth and Mars. Bound.-Layer Meteor., 184, 381–400, https://doi.org/10.1007/s10546-022-00713-w.

    • Search Google Scholar
    • Export Citation

  • Larotonda, J. M., and M. J. Townsend, 2021: Data release report for the Source Physics Experiment phase II: Dry Alluvium Geology experiments (DAG-1 through DAG-4), Nevada National Security Site. NNSS Rep. DOE/NV/03624-1220, 84 pp., https://www.osti.gov/servlets/purl/1825534.

  • Lorenz, R. D., and P. D. Lanagan, 2014: A barometric survey of dust-devil vortices on a desert playa. Bound.-Layer Meteor., 153, 555–568, https://doi.org/10.1007/s10546-014-9954-y.

    • Search Google Scholar
    • Export Citation

  • Lorenz, R. D., and D. Christie, 2015: Dust devil signatures in infrasound records of the International Monitoring System. Geophys. Res. Lett., 42, 2009–2014, https://doi.org/10.1002/2015GL063237.

    • Search Google Scholar
    • Export Citation

  • Lorenz, R. D., and B. K. Jackson, 2015: Dust devils and dustless vortices on a desert playa observed with surface pressure and solar flux logging. GeoResJ, 5, 1–11, https://doi.org/10.1016/j.grj.2014.11.002.

    • Search Google Scholar
    • Export Citation

  • Lorenz, R. D., and D. Reiss, 2015: Solar panel clearing events, dust devil tracks, and in-situ vortex detections on Mars. Icarus, 248, 162–164, https://doi.org/10.1016/j.icarus.2014.10.034.

    • Search Google Scholar
    • Export Citation

  • Lorenz, R. D., and B. K. Jackson, 2016: Dust devil populations and statistics. Space Sci. Rev., 203, 277–297, https://doi.org/10.1007/s11214-016-0277-9.

    • Search Google Scholar
    • Export Citation

  • Lorenz, R. D., L. D. Neakrase, and J. D. Anderson, 2015: In-situ measurement of dust devil activity at Ja Jornada Experimental Range, New Mexico, USA. Aeolian Res., 19, 183–194, https://doi.org/10.1016/j.aeolia.2015.01.012.

    • Search Google Scholar
    • Export Citation

  • Lorenz, R. D., and Coauthors, 2016: History and applications of dust devil studies. Space Sci. Rev., 203, 5–37, https://doi.org/10.1007/s11214-016-0239-2.

    • Search Google Scholar
    • Export Citation

  • Macpherson, K. A., J. R. Coffey, A. J. Witsil, D. Fee, S. Holtkamp, S. Dalton, H. McFarlin, and M. West, 2022: Ambient infrasound noise, station performance, and their relation to land cover across Alaska. Seismol. Res. Lett., 93, 2239–2258, https://doi.org/10.1785/0220210365.

    • Search Google Scholar
    • Export Citation

  • Merchant, B. J., 2015: Hyperion 5113/GP infrasound sensor evaluation. Sandia Rep. SAND2015-7075, 43 pp., https://doi.org/10.2172/1213302.

  • Mormon, S. A., and G. S. Plumlee, 2014: Dust and human health. Mineral Dust: A Key Player in the Earth System, P. Knippertz and J. B. Stuut, Eds., Springer, 385–410, https://doi.org/10.1007/978-94-017-8978-3_15.

  • Newman, C. E., H. Kahanp, M. I. Richardson, G. M. Martínez, A. Vicente-Retortillo, and M. T. Lemmon, 2019: MarsWRF convective vortex and dust devil predictions for Gale Crater over 3 Mars years and comparison with MSL-REMS observation. J. Geophys. Res. Planets, 124, 3442–3468, https://doi.org/10.1029/2019JE006082.

    • Search Google Scholar
    • Export Citation

  • Slinkard, M. E., D. B. Carr, and C. J. Young, 2013: Applying waveform correlation to three aftershock sequences. Bull. Seismol. Soc. Amer., 103, 675–693, https://doi.org/10.1785/0120120058.

    • Search Google Scholar
    • Export Citation

  • Snelson, C. M., R. E. Abbott, S. T. Broome, R. J. Mellors, H. J. Patton, A. J. Sussman, and W. R. Walter, 2013: Chemical explosion experiments to improve nuclear test monitoring. Eos, Trans. Amer. Geophys. Union, 94, 237–239, https://doi.org/10.1002/2013EO270002.

    • Search Google Scholar
    • Export Citation

  • Steakley, K., and J. Murphy, 2016: A year of convective vortex activity at Gale Crater. Icarus, 278, 180–193, https://doi.org/10.1016/j.icarus.2016.06.010.

    • Search Google Scholar
    • Export Citation

  • Sundermier, A., R. Tibi, R. A. Brogan, and C. J. Young, 2022: Applying waveform correlation to reduce seismic analyst workload due to repeating mining blasts. Bull. Seismol. Soc. Amer., 112, 171–190, https://doi.org/10.1785/0120210124.

    • Search Google Scholar
    • Export Citation

  • Tang, Y., Y. Han, and Z. Liu, 2018: Temporal and spatial characteristics of dust devils and their contribution to the aerosol budget in East Asia—An analysis using a new parameterization scheme for dust devils. Atmos. Environ., 182, 225–233, https://doi.org/10.1016/j.atmosenv.2018.03.050.

    • Search Google Scholar
    • Export Citation

  • Torrence, C., and G. P. Compo, 1998: A practical guide to wavelet analysis. Bull. Amer. Meteor. Soc., 79, 61–78, https://doi.org/10.1175/1520-0477(1998)079<0061:APGTWA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation

  • Virtanen, P., and Coauthors, 2020: SciPy 1.0: Fundamental algorithms for scientific computing in Python. Nat. Methods, 17, 261–272, https://doi.org/10.1038/s41592-019-0686-2.

    • Search Google Scholar
    • Export Citation

  • Walker, K. T., and M. A. Hedlin, 2010: A review of wind-noise reduction technologies. Infrasound Monitoring for Atmospheric Studies, Springer, 141–182.

  • <sc>Fig</sc>. 1.
    View in gallery
    Fig. 1.

    Microbarometer station locations (circles) and nearby weather station (triangle) in the Nevada National Security Site (denoted as a star in the overview map). Microbarometer stations used for background noise analyses are shown in red, and stations analyzed for vortex detections are shown in white.

  • <sc>Fig</sc>. 2.
    View in gallery
    Fig. 2.

    Convective vortex pressure signatures over a 1-h period on a subset of microbarometer stations (see Fig. 1) on 4 Jul 2018. Each spike, highlighted in gray, is consistent with the signature of a vortex overrunning a microbarometer. The signal has a characteristic “heartbeat” appearance due to the convolution of a pressure dip with the instrument response. Red scale bar denotes 30 Pa.

  • <sc>Fig</sc>. 3.
    View in gallery
    Fig. 3.

    Recorded (blue) and synthetic convective vortex (dashed red line) recorded on microbarometer station D10M0 at 1232 local time 4 Jul 2018, including (a) data prior to instrument response removal (blue) and synthetic generated from convolution of the instrument response with a Lorentzian vortex of 7-s FWHM and 60 Pa dip (dashed red). (b) Data following instrument response removal (blue) and synthetic Lorentzian profile from (a) before convolution with the instrument response.

  • <sc>Fig</sc>. 4.
    View in gallery
    Fig. 4.

    Histogram distributions from a week of data assessed over daylight hours from a variety of stations, including (a) Brownian profile constant, (b) Brownian profile slope, and (c) background noise levels. Note that the mean value is shown as a solid vertical lines and one standard deviation as dashed vertical lines for each distribution. (d) Power spectral density (PSD) of data from station D10M0 over 1700–2300 UTC (blue) and fit Brownian trend (magenta).

  • <sc>Fig</sc>. 5.
    View in gallery
    Fig. 5.

    Optimization of the wavelet detector for increased precision and recall. The original time series is plotted in blue, with red sections displaying the “detected” part of the time series. Red contours in the spectrogram indicate the region within which the signal is not produced by background noise with 95% confidence. Columns show time series (left) before and (right) after application of a given step. (a) Removal of instrument response leads to a reduction in the spurious amplification of noise relative to the vortex signal. (b) The wavelet spectrum is smoothed, removing localized pockets of noise-induced high wavelet power. (c) Speckles are rejected when the peak energy is above 0.1 Hz or the “detection” is shorter than 20 s in duration, based on observational data on naturally occurring pressure vortices. (d) Large vortices can influence the signal variance, causing the nondetection of smaller vortices in the same time series. A second pass with wavelet power scaled by outlier-removed variance allows for the detection of smaller vortices. A different time series was used for (d) compared to (a)–(c) to better illustrate the effect of that step.

  • <sc>Fig</sc>. 6.
    View in gallery
    Fig. 6.

    Record sections for a variety of convective vortex theoretical (a) Lorentzian and (b) heartbeat profiles arranged according to FWHM values, from low (bottom profiles) to high (top profiles), and shaded according to maximum absolute signal strength. These are the templates used in the correlation detector.

  • <sc>Fig</sc>. 7.
    View in gallery
    Fig. 7.

    Example 6-h synthetic seeded with heartbeat vortices and detections showing results from the (a) wavelet, (b) correlation, and (c) wavelet detector output as input for the correlation detector. The true detections are green windows for the wavelet detector and as overlaid templates and filled circles for correlation coefficient. False detections are red windows (wavelet detector) with a red × at the peak amplitude of the detection (wavelet and correlation detectors). Missed detections are shown as gray triangles.

  • <sc>Fig</sc>. 8.
    View in gallery
    Fig. 8.

    Performance metrics of detected and inserted convective vortices across the 10 000 six-hour synthetics, including (a) achievable recall of the wavelet detector, (b) achievable recall of the correlation detector, (c) precision of the correlation detector, (d) achievable recall of the combined wavelet and correlation detector, and (e) precision of the wavelet and correlation detector. Note that achievable recall is shown as a function of the inserted synthetic convective vortex FWHM and dip parameters, but precision is shown as a function of the detected FWHM parameter and correlation coefficient (CC).

  • <sc>Fig</sc>. 9.
    View in gallery
    Fig. 9.

    Day plot of time series pressure data from station D5M02 on 23 Jun 2018, with detection templates overlaid in gray.

  • <sc>Fig</sc>. 10.
    View in gallery
    Fig. 10.

    Summary of detection statistics over all analyzed stations in June and July, including (a) local time, (b) temperature, (c) humidity, and (d) FWHM of the detections, which contain increased detections with FWHM > 10 s due to current wavelet detector tuning.

Fig. 1.

Microbarometer station locations (circles) and nearby weather station (triangle) in the Nevada National Security Site (denoted as a star in the overview map). Microbarometer stations used for background noise analyses are shown in red, and stations analyzed for vortex detections are shown in white.
Fig. 2.

Convective vortex pressure signatures over a 1-h period on a subset of microbarometer stations (see Fig. 1) on 4 Jul 2018. Each spike, highlighted in gray, is consistent with the signature of a vortex overrunning a microbarometer. The signal has a characteristic “heartbeat” appearance due to the convolution of a pressure dip with the instrument response. Red scale bar denotes 30 Pa.
Fig. 3.

Recorded (blue) and synthetic convective vortex (dashed red line) recorded on microbarometer station D10M0 at 1232 local time 4 Jul 2018, including (a) data prior to instrument response removal (blue) and synthetic generated from convolution of the instrument response with a Lorentzian vortex of 7-s FWHM and 60 Pa dip (dashed red). (b) Data following instrument response removal (blue) and synthetic Lorentzian profile from (a) before convolution with the instrument response.
Fig. 4.

Histogram distributions from a week of data assessed over daylight hours from a variety of stations, including (a) Brownian profile constant, (b) Brownian profile slope, and (c) background noise levels. Note that the mean value is shown as a solid vertical lines and one standard deviation as dashed vertical lines for each distribution. (d) Power spectral density (PSD) of data from station D10M0 over 1700–2300 UTC (blue) and fit Brownian trend (magenta).
Fig. 5.

Optimization of the wavelet detector for increased precision and recall. The original time series is plotted in blue, with red sections displaying the “detected” part of the time series. Red contours in the spectrogram indicate the region within which the signal is not produced by background noise with 95% confidence. Columns show time series (left) before and (right) after application of a given step. (a) Removal of instrument response leads to a reduction in the spurious amplification of noise relative to the vortex signal. (b) The wavelet spectrum is smoothed, removing localized pockets of noise-induced high wavelet power. (c) Speckles are rejected when the peak energy is above 0.1 Hz or the “detection” is shorter than 20 s in duration, based on observational data on naturally occurring pressure vortices. (d) Large vortices can influence the signal variance, causing the nondetection of smaller vortices in the same time series. A second pass with wavelet power scaled by outlier-removed variance allows for the detection of smaller vortices. A different time series was used for (d) compared to (a)–(c) to better illustrate the effect of that step.
Fig. 6.

Record sections for a variety of convective vortex theoretical (a) Lorentzian and (b) heartbeat profiles arranged according to FWHM values, from low (bottom profiles) to high (top profiles), and shaded according to maximum absolute signal strength. These are the templates used in the correlation detector.
Fig. 7.

Example 6-h synthetic seeded with heartbeat vortices and detections showing results from the (a) wavelet, (b) correlation, and (c) wavelet detector output as input for the correlation detector. The true detections are green windows for the wavelet detector and as overlaid templates and filled circles for correlation coefficient. False detections are red windows (wavelet detector) with a red × at the peak amplitude of the detection (wavelet and correlation detectors). Missed detections are shown as gray triangles.
Fig. 8.

Performance metrics of detected and inserted convective vortices across the 10 000 six-hour synthetics, including (a) achievable recall of the wavelet detector, (b) achievable recall of the correlation detector, (c) precision of the correlation detector, (d) achievable recall of the combined wavelet and correlation detector, and (e) precision of the wavelet and correlation detector. Note that achievable recall is shown as a function of the inserted synthetic convective vortex FWHM and dip parameters, but precision is shown as a function of the detected FWHM parameter and correlation coefficient (CC).
Fig. 9.

Day plot of time series pressure data from station D5M02 on 23 Jun 2018, with detection templates overlaid in gray.
Fig. 10.

Summary of detection statistics over all analyzed stations in June and July, including (a) local time, (b) temperature, (c) humidity, and (d) FWHM of the detections, which contain increased detections with FWHM > 10 s due to current wavelet detector tuning.
All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 3490 2977 74
PDF Downloads 554 309 18