Volume 45, Issue 20 p. 10,903-10,911
Research Letter
Free Access

Coincidental TID Production by Tropospheric Weather During the August 2017 Total Solar Eclipse

Sebastijan Mrak

Corresponding Author

Sebastijan Mrak

Department of Electrical and Computer Engineering and Center for Space Physics, Boston University, Boston, MA, USA

Correspondence to: S. Mrak,

[email protected]

Search for more papers by this author
Joshua Semeter

Joshua Semeter

Department of Electrical and Computer Engineering and Center for Space Physics, Boston University, Boston, MA, USA

Search for more papers by this author
Y. Nishimura

Y. Nishimura

Department of Electrical and Computer Engineering and Center for Space Physics, Boston University, Boston, MA, USA

Search for more papers by this author
Michael Hirsch

Michael Hirsch

Department of Electrical and Computer Engineering and Center for Space Physics, Boston University, Boston, MA, USA

Search for more papers by this author
Nithin Sivadas

Nithin Sivadas

Department of Electrical and Computer Engineering and Center for Space Physics, Boston University, Boston, MA, USA

Search for more papers by this author
First published: 15 October 2018
Citations: 14


It has been proposed (Chimonas & Hines, 1970, https://doi.org/10.1029/JA075i004p00875) that a total solar eclipse should generate internal gravity waves (GWs) that manifest as traveling ionospheric disturbances (TIDs) at ionospheric heights. Zhang et al. (2017, https://doi.org/10.1002/2017GL076054) recently reported observations of electron density perturbations trailing the region of maximum obscuration, claiming the results as the first unambiguous evidences for eclipse-induced bow waves. We present evidence showing extensive TID activity on two consecutive days, the day of the eclipse and the day before. A particularly intense TID concentric wavefield emerged from the background ionosphere 5 hr before the arrival of the totality and persisted there throughout the eclipse. The apparent center was located over Iowa/South Dakota region, 300–500 km north from the eclipse path. We examine concurrent observations of tropospheric and ionospheric weather and find a great spatiotemporal correlation. TID wave parameters do agree with previous observations and models of thunderstorm-generated GWs/TIDs; conversely, the wave parameters are an order of magnitude off from modeling results for eclipse-generated GWs/TIDs.

Key Points

  • The TIDs previously ascribed to the 2017 total solar eclipse are not associated with the eclipse; they were present at other times
  • The TID pattern, speed, and wavelength are inconsistent with the bow-wave hypothesis; evidence is presented, suggesting a thunderstorm origin
  • The TIDs with a concentric pattern propagated radially away from the point source in the east/southeast direction

Plain Language Summary

Recent studies of ionospheric dynamics on the day of the 2017 total solar eclipse observed concentric waves in the ionosphere and attributed them to the eclipse itself. We present a concurrent analysis of ionospheric and tropospheric weather and find a strong correlation between thunderstorm activity and the ionospheric wavefield. We find that the ionospheric waves emerged 5 hr before the eclipse and they persisted within a confined region of space throughout the end of the eclipse. A careful analysis of these waves shows remarkable agreement with previous studies of tropospheric weather-generated gravity waves and their imprint in the ionosphere as traveling ionospheric disturbances. Further comparison with models and other recent observations of the 2017 total solar eclipse further supports our conclusions to ascribe the ionospheric wavefield to the tropospheric weather as the driving mechanism. In aggregate, the ionosphere on the day of the eclipse provided us with an intriguing pattern, showing simultaneous forcing from the Sun and from the troposphere.

1 Introduction

A total solar eclipse is an episodic natural laboratory experiment that provides an incredible opportunity to study the response of the thermosphere, ionosphere, and atmosphere (geospace) to a controlled perturbation. The controlled nature of the experiment is invaluable for testing and validation of physical models. It is therefore crucial to fully understand the observations, including the influence of multiple source mechanisms that may be acting on the system (e.g., Coster et al., 2017; Mrak et al., 2018). Of particular interest in the ionospheric community has been a proposed initiation of atmospheric bow waves (Chimonas & Hines, 1970). The rise of GPS receiver networks has allowed us to map ionospheric features in space and time. The abundance of GPS receivers in North America and the ability to receive multiple lines of sight simultaneously define an opportunistic ionospheric imaging tool with unprecedented resolution. A differential total electron content (ΔTEC) approach has been successfully employed to detect tiny spatially coherent disturbances caused by numerous physical mechanisms (e.g., Azeem et al., 2015; Chou et al., 2017; Grawe & Makela, 2015; Nishioka et al., 2013; Tsugawa et al., 2007, 2011).

Hitherto, there are two known reports of possible bow waves imprinted in the ionosphere (i.e., Liu et al., 2011; Zhang et al., 2017). However, both reports rest on weak or questionable foundations. While the former (Liu et al., 2011) suffers from sparse spatial sampling (averaging 500 km in longitude), the latter (Zhang et al., 2017) neglected the background space weather during and prior to the eclipse. In this letter, we carefully analyze ionospheric electron density perturbations, ΔTEC, observed before and during the 2017 total solar eclipse and find no evidence connecting their generation to the eclipse.

Observations of dayside ionospheric activity for two consecutive days revealed a zoo of ionospheric perturbations, among which the most prominent were concentric traveling ionospheric disturbances (TIDs) propagating from central United States toward east/southeast. The spatiotemporal evolution and wave parameters of these TIDs closely match the bow waves reported by Zhang et al. (2017). Using an extended set of ionospheric observations in conjunction with radar maps of weather reflectivity, we show that the TID source is likely to be internal gravity waves (GWs) initiated by a region of persistent thunderstorm activity (convective plumes). The TIDs were, in fact, present in the background ionosphere before the arrival of the eclipse. Additionally, there was very similar TID activity over the same region in space and time on the day before. The source of the TIDs is linked to a tropospheric weather system by virtue of space-time-frequency wave analysis and simultaneous observations of ionospheric and tropospheric weather. Based upon the data analysis, we argue that the TIDs on the day of the eclipse were initiated by tropospheric weather rather than the eclipse. We also elaborate on the observational and physical difficulties in linking the observed TIDs to an eclipse source.

2 Methodology

We use Continuously Operating Reference Station and Crustal Dynamics Data Information System publicly available databases with Global Navigation Satellite Systems (GNSS) data, which totally account for ∼1,800 receivers in the continental United States. We utilized a standard approach to compute the phase-corrected slant TEC estimates (Coster et al., 1992), converted to vertical (vTEC) via mapping function (Klobuchar, 1987), applied at 300-km altitude. We then subtract the background vTEC to obtain differential TEC (ΔTEC) residuals, utilizing variable orders of polynomials (Mrak et al., 2018). The carrier-phase-based differential approach provides accuracy better to 0.03 TECu (Coster et al., 2012; 1 TECu = 1016e/m2). The ΔTEC residuals are then mapped to a geographical map at an altitude of 300 km and then transformed from the naturally irregular spatial grid into a regular grid (e.g., Azeem et al., 2015; Mrak et al., 2018; Nykiel et al., 2017) with a resolution 0.2° × 0.2° (geographical coordinates). The differential approach and excellent spatial coverage (∼15,000 spatial points at a given time) allow one to extract coherent spatial features of tiny amplitudes. The spatial extent and appearance of the coherent perturbations are then presented in form of 2-D projections. The wave parameters are then extracted using keogram and spectrogram analyses.

3 Observations

Figure 1 shows TEC perturbations on the day of the total solar eclipse (21 August 2017). Starting at 13 UT, a coherent TID wavefield emerged over central and eastern United States with longitudinal extent ranging from Iowa to the Atlantic and latitudinal extent from the Gulf of Mexico to Canada. The TIDs were nearly concentric, with an apparent center over the Iowa/South Dakota region. The TIDs were the dominant coherent dayside ionospheric perturbation until the arrival of the eclipse (penumbra) at ∼17 UT. During the eclipse, the predominant ionospheric perturbations were related to irregular extreme ultraviolet illumination caused by two solar active regions (cf. Mrak et al., 2018). Nevertheless, the presence of the background TIDs is apparent and appears as a modulation, superposed on the eclipse modification. The TIDs were apparent in the same spatial domain, as they were prior to the eclipse. Figure 2 shows a snapshot at 18:42 UT, when the totality was already exiting the continental United States. The TID wavefield is clearly visible as a superposed modulation over the dominant eclipse modification. Figure 2b encompasses merged ΔTEC map and tropospheric weather map, where the red X mark denotes a region of most intense precipitation (reflectivity ≥ 60 dBZ) and dashed fiducial circles bolster the apparent concentric figure, with a center in the thunderstorm system.

Details are in the caption following the image
(a–h) 21 August 2017: Differential TEC ΔTEC images, mapped at 300-km altitude, showing extensive traveling ionospheric disturbance activity over central and eastern United States. The arrows in (b) indicate propagation direction of underlying traveling ionospheric disturbances. The black fiducial dashed line in (e)–(h) indicates the path of totality. (e)–(g) Dashed lines indicate positions of large-scale TEC perturbations due to uneven EUV flux. (e) White circle indicates the position of the total eclipse at the given time. TEC = total electron content; EUV = extreme ultraviolet.
Details are in the caption following the image
(a) An image of traveling ionospheric disturbances trailing the totality (white circle), the TIDs are modulated on top of the salient eclipse-inducted modification. The image bolsters simultaneous and cooperative forcing of the ionosphere from below (TIDs) and from above (large-scale TEC perturbations). (b) Same as panel (a), with an overlay of tropospheric weather storms (gray) from the Next Generation Weather Radar maps. Red X denotes a position of the most intense precipitation inside the storm system. Dashed fiducial lines emphasize concentric nature of the TIDs, with a center in the storm system, center of the red X. (c) Time series plot of ΔTEC perturbations for three regions identified X in panel (a). (d) A representative spectrogram of ΔTEC at 90°W, 40°N. TIDs = traveling ionospheric disturbances; TEC = total electron content.

The wave period was computed for three locations, marked as colored X marks in Figure 2a, utilizing the time-dependent Fourier transform with a fully overlapping 2-hr window and the Hamming window function. Figure 2c shows example ΔTEC time series for the three locations, and Figure 2d shows a representative spectrogram at the midpoint (90°W, 40°N, red X). The spectrogram shows TID frequency range, with persistent components clustered in a range from 0.5 to 1 mHz (periods Tp = 16.6–33.3 min) between 13 and 20 UT. The peak in the TID waves is centered at ∼0.75 mHz, which corresponds to a wave period Tp = 22 min. The most prominent wave activity, however, is centered at 0.35 mHz (Tp = 55 min), which was caused by the eclipse-inducted EUV modulation (Mrak et al., 2018).

Figure 3 shows contemporary maps of tropospheric reflectivity as observed by the Next Generation Weather Radar. Comparing with Figures 1 and 2, there is a notable spatiotemporal correlation between the TIDs and thunderstorm activity. At the beginning (∼13 UT), the apparent center of TIDs was near 95°W, 43°N (Iowa), and they were actively generated up until ∼14 UT when the causative thunderstorm broke apart. The

chronological progress of the TID wavefield is depicted in a longitudinal keogram in Figure 4a. After the initial thunderstorm collapsed, the western edge of TID wavefield slowly moved toward east, until subsequent wavefronts emerged further west at ∼15:30 UT (Figure 4a and Movie S1 in the supporting information), this time with an apparent center close to 100°W, 44°N (South Dakota). The resulting TIDs are seen in ΔTEC maps and keogram analysis in Figure 4. The TID wavefield resided there up until ∼19 UT. A similar pattern is seen in the Next Generation Weather Radar reflectivity maps (Figure 3 and Movie S2); until ∼14 UT, there was a strong, almost stationary thunderstorm over Iowa (Figures 3a and 3b); then the thunderstorm slowly disappeared, and a new thunderstorm system strengthened over South Dakota (Figures 3c and 3d), which lasted then for hours. We indicate a region of biggest reflectivity on each panel with red X. Most storm systems, however, had a reflectivity value >20 dBZ, which can be, in general, enough to produce deep convective plumes (Vadas et al., 2012). We direct the reader to watch Movies S1 and S2, showing ionospheric and tropospheric weather evolution, respectively.

Details are in the caption following the image
(a–d) Next Generation Weather Radar (NEXRAD) radar reflectivity maps of thunderstorm activity for the day of the eclipse. Each image shows an identified center (red cross X) of the most intense thunderstorm. The NEXRAD maps were produced by NEXRAD-quick-plot software (Hirsch, 2018).
Details are in the caption following the image
A set of traveling ionospheric disturbance keograms on the day of the eclipse. (a) Longitudinal cut at 40°N and (b) latitudinal cut at 90°W. (a) Solid black fiducial line is a local sunrise terminator delayed for 90 min; dashed fiducial line is a line of the biggest obscuration at the chosen latitude; black box is a region shown in panel (c).

We examine the TEC perturbations in greater detail here. Figure 4 consists of two keograms oriented along longitude and latitude (a and b) for the day of the eclipse. The TIDs emerged at ∼13 UT with a longitudinal extent between 100° and 70°W (edge of the GNSS coverage); the line of emanation is well aligned with a 90-min delayed local sunrise time (Figure 4a). The longitudinal keogram elongated along 40°N latitude shows a fairly constant slope (zonal velocity vx) of the TIDs as a function of longitude and time. The only notable modification was at ∼16 UT, when the slope slightly increased. The zonal velocity vx at 40°N latitude changed from ∼140 to ∼180 m/s; the time of the change is aligned with the new front of TIDs with a center of the second thunderstorm system. The latitudinal keogram (Figure 4b) shows the TIDs over the same time period with bent wavefronts. The wavefronts are vertical at ∼43N (13–14 UT), and the slope northward and southward is bent in opposite propagation directions. The wavefronts' shape implies the source/center was located near ∼43N at that time. The mean meridional velocity vy of the TIDs, extracted from the keogram (Figure 4b), is ∼170 m/s before and ∼140 m/s after the change. Therefore, the horizontal propagation velocity vh of the TIDs is estimated to be ∼220 m/s, propagating radially away from the source. The change of velocity components over time resulted in a minor change in the speed: from 220 m/s at the beginning to 230 m/s after the modification. The observed change in velocity components suggests its source had moved. In addition, we show a closer look at the TIDs during the eclipse time period in Figure 4c. It is nicely seen that the large-scale perturbations were modulated with a secondary wavefield. While the modulation effect is barely seen near the region of maximum obscuration (∼100W), the modulation is becoming evident in the direction eastward toward regions of smaller eclipse effect. The modulation is, however, colinear with the TIDs existing before the arrival of the eclipse.

We performed a 2-D Fourier analysis of the keograms (Figures 4a and 4b), to determine the dominant wave numbers of the TEC perturbations for the day of the eclipse. The spectrograms in Figure 5 show persistent waves, with wave numbers clustered in ranges kx = [0.025, 0.08] km−1 in the zonal direction and kx = [0.015, 0.04] km−1 in the meridional direction. The corresponding wavelength components are λy = [160, 420] km and λx = [80, 250] km. A mean horizontal wavelength λh is ∼350 km. However, it is impossible to retrieve the horizontal wavelength components of concentric waves in a nonradial direction. Therefore, we name ranges of existing wavelength components along single meridional and zonal lines, retrieved from spectrograms in Figure 5. The center of the concentric TIDs was not stationary; therefore, a keogram analysis in the thunderstorm frame of reference would be convoluted.

Details are in the caption following the image
Spectrograms of the keograms (Figures 4a and 4b, respectively. (a) Wave number decomposition in the zonal direction (kx). (b) Wave number decomposition in the meridional direction (ky). The red fiducial lines indicate a time range of the eclipse present in a field of view. Intensity of the spectrogram is TID power spectra in linear scale. TID = traveling ionospheric disturbance.

In addition, ionospheric electron density perturbations on the eclipse day was strikingly similar to the day before (20 August 2017). An illustrative example is depicted in Figure 6, where ΔTID map has overlaid radar reflectivity map. Fiducial circles indicate concentric nature of the observed TIDs, again having a center in the storm system. The TIDs on that day resided over nearly the same region in space, while the timing of the TIDs' appearance was also nearly synchronized. The TIDs emerged from the background ionosphere at ∼13 UT, which is about 1.5 hr after local sunrise (sunrise was at ∼11:30 UT at 90°W, 40°N). The TIDs also emerged everywhere at the same time, implying that the source of the TEC perturbations was present earlier, but their imprint in the ionosphere was revealed only after ionospheric production initiated. Movies S3 and S4 show TID and thunderstorm activity on that day, respectively.

Details are in the caption following the image
A snapshot of contemporary images of ionospheric (ΔTEC) and tropospheric (NEXRAD) weather maps on 20 August 2017 (the day prior to the eclipse). The map is a snapshot taken at 14:00 UT, ΔTEC map (color coded) is projected to 300-km altitude, and the NEXRAD maps are gray shaded. Red mark X is a region of biggest reflectivity (check Movie S4), and dashed circles match the traveling ionospheric disturbance pattern and have center in red X. TEC = total electron content; NEXRAD = Next Generation Weather Radar.
The obtained wave parameters are in agreement with previous studies of thunderstorm-induced/weather-induced GWs (e.g., Azeem et al., 2015; Azeem & Barlage, 2017; Chou et al., 2017). Furthermore, the estimated vertical wavelength λz for no background wind, taking the mean apparent horizontal wavelength λh = 350 km and buoyancy period Tb = 10.5 min, at 250-km altitude from the NRLMSISE-00 model is
which is enough to cause a prominent compression/ratification of the ionosphere in the vicinity of the F peak, as previously identified by Azeem and Barlage (2017). The horizontal angle Θ of GWs propagation was
Utilizing a simplified expression (Vadas et al., 2012) for vertical group velocity, cgz, we obtain the following:

The propagation time for these GWs to reach 250-km altitude is thus ∼36 min, which implies there is a nonnegligible time lag between tropospheric and ionospheric weather. It should be noted that a propagation time is usually longer (40–90 min), utilizing a GW dispersion relation (Yue et al., 2009).

4 Discussion

We present extended observations of ionospheric perturbations on the day of the total solar eclipse. We show concentric/elliptical TIDs, which resided over central United States; they propagated in east/southeast direction and had an apparent center in a causative thunderstorm system. The TIDs resided in the same position in space for a time period of ∼6 hr. The TIDs had a zonal wavelength range λx = [80, 250] km and meridional wavelength range λy = [160, 420] km, with a dominant horizontal wavelength λh≈ 350 km. The propagation speed was ∼220 m/s radially away from the source, predominantly in east/southeast direction. Further, estimated TID wave period range of the TIDs at 90°W, 40°N was Tp = [16.6, 23.8] min, with an estimated mean value of Tp≈ 22 min.

Estimated TID parameters are well within the range of previously reported tropospheric-weather-initiated GWs and/or subsequent TIDs (e.g., Azeem & Barlage, 2017; Azeem et al., 2015; Chou et al., 2017; Lay et al., 2013; Vadas & Nicolls, 2008; Vadas et al., 2012; Yue et al., 2009), as well as the ones reported by Zhang et al. (2017). The eclipse, however, did appear to alter the dominant wavelength and speed of the TIDs. This observation is a straightforward consequence of eclipse-induced erosion of the E region. The peak momentum flux carried by a GW is at an altitude between 150 and 200 km at solar minimum (Vadas & Fritts, 2006). Therefore, the erosion of the bottomside ionosphere raised the GW-TID coupling altitude to a region of higher temperature. The resulting TIDs within the eclipse should thus appear with larger horizontal phase speed and increased wavelengths (Vadas, 2007; Vadas & Fritts, 2006). Namely, the eclipse affected the TID wavefield; however, the effect was the one of modulation rather than the one of production.

Another eclipse modification can be seen as a temporary disappearance of the TIDs in the leading half of the penumbra. Yue et al. (2009) did a climatological study of concentric GWs, caused by tropospheric weather, and found that they occur only at times of the weakest neutral wind (May–June and August–September, with winds weaker than 20 m/s) in the mesosphere/stratosphere. Additionally, Harding et al. (2018) showed a huge wind modification carried by the penumbra, with the wind direction pointing toward the totality and with winds exceeding 50 m/s right in front of the total obscuration. Therefore, the eclipse-imposed winds in that area damped the GWs, since the imposed modification is at least a factor 2 bigger than the background winds; hence, no TIDs were observed there.

Considering the extended set of observations, including ionospheric TIDs and the weather activity fully covered in Movies S1S4, we argue that the TIDs trailing the totality were not initiated by the eclipse. Instead, they resemble the same source field as prior to the eclipse. If the waves would have been triggered by the eclipse, then the TIDs should be apparent along its entire path. Instead, the waves were spatially confined only to the region of the prior TID wavefield. The apparent concentric/elliptical shape and the source region indeed coincide with the path of the totality, which can be a deceiving factor. Nonetheless, the actual center of the TIDs was slightly northward from the totality track (∼3 latitude), which breaks the symmetry that would be expected if the source was the eclipse. Further, the center of the TIDs (Figure 2) spatially matched the center of the thunderstorm over South Dakota (Figures 3c and 3d).

Another observation that deserves discussion is the timing of the waves apparent behind the totality region. As we calculated in equation3, the apparent waves propagate vertically with a speed ∼110 m/s. If the source would be the stratospheric cooling (Chimonas & Hines, 1970), then it would take Δtz/cgz≈33 min to reach 250-km altitude. However, the TIDs trailing the totality became apparent, ∼10 min behind the totality, namely, close to a region where the neutral winds abate and change the direction (cf. Harding et al., 2018). If the source of the TIDs were the eclipse, the source altitude would reside at Δztcgz≈66 km, at an altitude of 184 km (250–66 km).

The bow waves' interpretation is also problematic on theoretical grounds. Both Liu et al. (2011) and Zhang et al. (2017) linked their observations to the original prediction by Chimonas and Hines (1970) and subsequent steady state analysis (Chimonas, 1970). But at ionospheric heights, the bow waves anticipated by Chimonas (1970) should have a wavelength of ∼1,000 km and reside at a lateral distance of ∼10,000 km away from the totality track. A more thorough calculation by Fritts and Luo (1993) further confirmed these parameters, with results in the same magnitude range as obtained by Chimonas (1970). Even a more recent treatment of the same problem, invoking a comprehensive first-principles analysis (Eckermann et al., 2007), confirmed these predictions. In summary, the physics-based modeling efforts do predict bow waves of the stratospheric origin at ionospheric heights. However, the waves should have the following parameters: (i) horizontal wavelength ∼1,000 km and (ii) lateral extent of ∼10,000 km; and (iii) the waves should lag the totality by ∼1 hr (Eckermann et al., 2007). The observed TIDs also claimed as the bow waves (Zhang et al., 2017) are in stark disagreement with theoretical predictions, adding further support for the alternate source mechanism we have identified.

Curiously, there are recent studies that confirmed the existence of the eclipse-generated bow wave in the thermosphere for this eclipse. Namely, Harding et al. (2018) showed the observational evidences for thermospheric bow wave, while Lin et al. (2018) and Lei et al. (2018) showed modeled thermospheric/ionospheric bow waves. However, it should be noted that the latter authors observed an evanescent in situ generated bow wave (Ridley et al., 1984) and not a wavefield generated at stratospheric heights.

5 Summary

The geospace is an extremely complicated and entangled system, which demands a careful examination, even if perturbed by a controlled experiment. We demonstrated the complexity of systematic deconvolution of integrated and simultaneous driving of the Earth's ionosphere. The eclipse provided a marvelous sign of a pinhole projection of solar active regions and a simultaneous projection of tropospheric weather, as we demonstrated here. In this letter, we discussed the most likely origin of the accompanying traveling concentric ionospheric disturbances by virtue of concurrent observations of tropospheric and ionospheric weather. We found the presence of TID waves that emerged 5 hr before the arrival of the eclipse and that persisted throughout the penumbra. We find the following:
  1. The apparent centers of the TIDs coincided with locations of tropospheric thunderstorms.
  2. TID wave analysis revealed consistency with previously described tropospheric weather-initiated GWs and subsequent TIDs.
  3. The coincidental appearance of the TIDs can be explained by means of the eclipse timing, which happened in the month of August: a month of weakest neutral winds that allow GWs to propagate into thermosphere/ionosphere.
  4. Based upon a consideration of other observations and modeling results, we ruled out the eclipse source as a possible physical source of the observed concentric TIDs.

In summary, we have observed a myriad of unexpected and intriguing ionospheric phenomena during the August 2017 total solar eclipse, owing to the best remote sensor coverage in the history of eclipse observations. We report extensive TID activity on that day, which was most likely the ionospheric manifestation of thunderstorm-initiated GWs and not produced by the eclipse. The findings are important because there is an extensive eclipse modeling effort ongoing for this event as well as forthcoming total solar eclipses. We argue that without consideration of the tropospheric convective storms at the locations identified herein, one will not be able to reproduce the observed TIDs.


This work was supported by the National Science Foundation under grant AGS-1743832. The TEC maps were created using data from the CORS database publicly available on ftp://geodesy.noaa.gov/cors/ and CDDIS database ftp://ftp.cddis.eosdis.nasa.gov/gps/data/daily/. Totality track was obtained from https://eclipse.gsfc.nasa.gov/SEpath/SEpath2001/SE2017Aug21Tpath.html. NEXRAD radar data were acquired through Hirsch (2018), using a publicly available database: https://mesonet.agron.iastate.edu/archive/data/.