Total electron content (TEC) measured by Global Positioning System (GPS) receivers in the United States Great Plains is examined for three nights with large thunderstorms and for one night with little thunderstorm activity. The GPS TEC data are fit with a polynomial, and the variations are estimated by subtracting this fit from the data. We found that anomalous TEC variations are closely associated in time and space to the large underlying thunderstorms. The largest storm-related TEC variation is observed to be ~1.4 total electron content unit (TECU) over a typical nighttime background value of several TECUs. The variations near the storm appear to have more high-frequency content than those away from the storm, with periods of minutes to tens of minutes. No detectable localized TEC variation is observed for the thunderstorm-quiet night.
 Variation in ionospheric plasma has traditionally been attributed to changes in solar radiation and geomagnetic activity, which represent the key drivers in most empirical and assimilative models [e.g., Bilitza et al., 2011; Hajj et al., 2004]. However, other observations indicate gravity wave effects on the ionosphere from tropical and midlatitude storms [e.g., Bishop et al., 2006; Kelley, 1997]. More recently, the ionospheric community has begun to realize that tropospheric weather (below ~12 km) could have a significant effect on ionospheric electron distribution [e.g., Immel et al., 2009; Vadas and Fritts, 2004]. Specifically, a synoptical monthlong average of 135.6 nm emissions from the IMAGE-far ultraviolet Sensor found enhanced electron and O+ densities that were consistent with nonmigrating diurnal tides driven mainly by tropospheric weather [Immel et al., 2009]. In addition, modeling of atmospheric gravity waves (AGWs), originating from thunderstorms, has predicted variations in total electron content (TEC) associated with these AGWs of ±7% [Vadas and Liu, 2009].
 A statistical study by Davis and Johnson  found a correlation between intensification in the sporadic E layer and lightning activity. Our recent studies on the D layer ionosphere (~65–90 km altitude) have shown that AGWs originating from large mesoscale thunderstorms clearly perturb the electron distribution at the lower boundary of the ionosphere [Lay and Shao, 2011a], and electrical activity within even a small storm affects the D layer above by heating electrons and consequently reducing the density of the free electrons [Shao et al., 2013]. In this paper, we extend the observations vertically through the entire ionosphere by analyzing TEC observations from an array of ground-based Global Positioning System (GPS) receivers near large mesoscale thunderstorms.
2 Data Analysis
 The TEC measurements presented are from several GPS receiving stations of the Continuously Operating Reference Station (http://geodesy.noaa.gov/CORS/) network for the nights of 22–24 May 2005 and 17 June 2005, receiving the L1 and L2 GPS signals (1.58 and 1.22 GHz, respectively). The GPS receivers detect the integrated electron density along a line of sight from satellite to receiver (slant TEC) by measuring the dispersive contributions to the GPS L1 and L2 satellite signals. As is standard practice, TEC is determined by a combination of the rough but absolute pseudorange measurement and the precise but ambiguous phase differential measurement (phase TEC). The pseudorange measurement accounts for satellite and receiver biases and has an accuracy of 2–4 total electron content unit (TECU). In order to sensitively detect TEC perturbations, the TEC is detrended by subtracting a best fit polynomial. This process removes the gross offset, and the detrended result is mostly related to the phase TEC (0.01–0.1 TECU accuracy) [Burrell et al., 2009]. The TEC processing was performed using a software developed at Boston College [Seemala and Valladares, 2011], which differences the phase and code values at the L1 and L2 frequencies. The slant TEC measurements have been converted to vertical TEC (VTEC) values by dividing them by the so-called single layer mapping function, where Re is the Earth radius, ε is the satellite elevation angle, and h is the ionospheric height, assumed to be 350 km.
 The time period analyzed for this study is 02:00–06:00 UTC (20:00–00:00 local time, LT). All three nights analyzed in May 2005 have low (Kp < 4) geomagnetic activity. May 22 has the highest Kp of 2.0–3.7. May 23 has Kp of 0.7–1.7, and May 24 has Kp of 0.0–0.7. June 17 2005 has a maximum Kp of 4.7, indicating medium-high geomagnetic activity.
 The “undisturbed” ionospheric VTEC is calculated by fitting a 6th order polynomial to the VTEC measurement from each station, similar to the method described in Ozeki and Heki . Figure 1 shows examples of the VTEC measurements and corresponding polynomial fits on 23 May 2005 for stations vcio (Figure 1a) and lthm (Figure 1b) detecting signals from satellite with the given GPS pseudo-random number (PRN) 13. In Figure 1a, the measured VTEC deviates very little from the fitted curve, while in Figure 1b, there are larger deviations between ~03:30 and 05:00 UTC. The deviation of the VTEC from “undisturbed” conditions is obtained by subtracting the best fit polynomial from the VTEC measurement. The removal of this fit acts to filter out lower-frequency variations, so effectively this method is only sensitive to higher-frequency variations.
 The top row of Figures 2a, 2c, and 2e shows deviations in VTEC from “undisturbed” conditions on 22–24 May 2005 for several stations detecting the GPS signal from satellite PRN 13. Results from each station are vertically offset from the previous station by 1 TECU (1016 electrons/m2). The bottom row of Figures 2b, 2d, and 2f shows the corresponding paths of the ionospheric pierce points (IPPs) at 200 km for the corresponding satellite-receiver lines-of-sight. The altitude of 200 km was chosen for the IPPs under the assumption that fluctuations originating from tropospheric phenomena will be lower than the typical 350 km peak of the ionospheric electron profile, while the 350 km height was used in the mapping function to account for the bulk plasma. As an example, AGWs modeled in Vadas and Liu  often dissipate below 200 km. In this paper, lightning activity is used to represent the intensity of the parental storms, so the maps of Figures 2b, 2d, and 2f also show underlying lightning events detected between 00:00 UTC and 06:00 UTC (18:00–00:00 LT) with a density plot in gray scale. The contour levels represent 0.5 (lightest gray), 2.5, 4.5, 6.5, and 8.5 (black) km−2. Regions with nonzero lightning density but less than 0.5 km−2 are bounded by the light blue contour line. The lightning events were detected by the Los Alamos Sferic Array [Shao et al., 2006]. The color in all panels of Figure 2 indicates UTC time.
 The night of 22 May (first column) has very little lightning activity: the blue contour line shows regions with nonzero lightning activity but with a maximum lightning density of less than 0.5 km−2. The red “x” in Figure 2b shows the main location of lightning activity during the peak lightning rate of 42 per minute at 00:40 UTC. The black “x” shows a lightning region with a peak lightning rate of 6 per minute at 01:45 UT and a second peak of 25 per minute at 06:00 UT.
 May 23 (middle column) has a mesoscale storm with extreme lightning activity (>8.5 km−2) in a fairly concentrated area. The lightning rate is relatively low (<10 per minute) before 03:15 UTC, but then quickly increases to a maximum of 920 per minute at 06:00 UTC. The red “x” in Figure 2d indicates the main location of the storm at 03:15 UTC when intense electrification and updraft activity caused the lightning rate to increase rapidly. Earlier observations of this same storm showed fluctuations at the lower boundary of the ionosphere that were suggestive of thunderstorm-produced AGW effects [Lay and Shao, 2011b]. The technique in Lay and Shao [2011a, 2011b] for D layer measurement is based on lightning-produced radio signals, and the spatial and temporal resolution of the measurement depends on the lightning density and rate.
 The night of 24 May (last column) also has intense lightning activity, but with a maximum lightning density of only 4.6 km−2. Maximum lightning rates in the storm are ~380 per minute at 03:00 UTC (location shown by red “x”) and again at 06:00 UTC (black “x”). In general, lightning rates for this mesoscale storm were greater than 100 per minute for the entire time from before 00:00 UTC to after 06:00 UTC. Because the storm on 24 May 2005 has lower lightning rates and densities, it is difficult to apply the technique described in Lay and Shao [2011a, 2011b] for a high-resolution D layer measurement.
 Figure 2a shows that TEC fluctuations on 22 May are small in amplitude and mainly uniform among all the stations over the entire region, indicating that there are no clear localized fluctuations originating within this region. On 23 May however larger TEC variations are clustered to the east and north of the storm (Figure 2c). The largest deviations occur at stations lthm, cnwm, and blmm, while Figure 2d shows that the paths for those stations pass closely to the east and north of the storm. Path slai also passes just north of the storm and shows a smaller, but clear variation as well. Path nds1 passes directly over the storm and shows a slight variation at ~05:00 UTC when it is just past the storm to the south. The remaining paths are all west and south of the storm and show no clear variation during the times when the measurement paths are near the large thunderstorm. Paths oktu and okhv pass very close to the storm region, but earlier in the night (02:00–03:30 UTC) before any significant thunderstorm activity. The peak-to-peak magnitude of the variation on path lthm is about 0.75 TECU. This is a large variation relative to the total minimum VTEC of ~2.5 ± 2–4 TECU (Figure 1b) during this nighttime measurement. In comparison, Vadas and Liu  reported 5% TEC variation due to AGWs generated by a convective plume 1–1.5 h before sunset. Based on the International Reference Ionosphere [Bilitza et al., 2011], a typical pre-sunset background of ~40 TECU is inferred for the general geographical location. Therefore, a 5% change implies a 2 TECU variation. While the result inferred from Vadas and Liu (2 TECU) and that in this paper (0.75 TECU) could be considered comparable, the pre-sunset results cannot necessarily be extrapolated to the nighttime ionosphere in comparing fluctuation amplitudes. In addition to the larger fluctuation amplitudes along paths blmm, cnwm, slai, lthm, and nds1 compared to other paths, those variations also seem to have more high-frequency content than that of the general fluctuations in background activity, such as in Figure 2a.
 Analysis of TEC fluctuations on 24 May 2005 (Figures 2e and 2f) shows similar results to the findings near the storm on 23 May. A localized, high-frequency perturbation is evident on paths near the storm (mbww, wylc, pltc, gdac, amc2) and not on other paths. In this case, the perturbation appears to be to the north/west of the storm instead of north/east. Although not shown, paths farther west (at ~110° longitude) of the plotted region show no clear variation.
 Figure 3 shows similar variations on 23 May 2005 detected from a different satellite, PRN 23. In this case, the fluctuations north and east of the storm are even more pronounced than in the PRN 13 case, with a maximum peak-to-peak variation of ~1.4 TECU on path cnwm. Consistent with Figures 2c and 2e, these enhanced fluctuations seem to have more high-frequency content. Results for 22 May and 24 May from PRN 23 are also consistent with the data shown in Figure 2 and are not shown here. These two satellites (PRN 13 and 23) pass over this region during the time period of the fluctuations (~03:00–05:00 UTC) with maximum elevation angle of greater than 75°. Satellites PRN 3 and 19 also pass over this region during this time, but with maximum elevation angles of ~55° and ~60°, respectively. TEC variations are also visible from these satellite links in the same localized area, but are not as pronounced, possibly due to the longer slant path of integrated electron density. This would average over more features in regions farther away from the storm and dilute any signature from the storm.
 Finally, Figure 4 shows results from 17 June 2005 for satellite PRN 19. Although this day has medium-high geomagnetic activity (Kp = 4.7), it also has a very active mesoscale thunderstorm that was previously observed to produce D layer fluctuations propagating outward from the storm [Lay and Shao, 2011a]. This storm has sustained lightning rates of >700 per minute from 00:00 UTC until 04:00 UTC, and then the rate decreased slowly to 400 per minute at 06:00 UTC. The maximum lightning density was 13.5 km−2 at −99° longitude, 37° latitude. Because of the higher geomagnetic activity than on 22–24 May (Kp: 3.7, 1.7, 0.7, respectively), the background variation of TEC is larger in Figure 4a. However, Figure 4a shows small-amplitude but very high-frequency fluctuations superimposed on those variations at stations txam, gdac, tcun, nmro, and nmsf, all of which are south and west of the storm. Station jtnt also detects high-frequency fluctuations between 03:30 and 04:00 UTC, during which time, the path lies atop the storm. These fluctuations have periods of ~4 min.
4 Summary and Discussion
 The nights of 23 May, 24 May, and 17 June 2005 have significant thunderstorm activity that corresponds well in time and space to localized regions of enhanced TEC fluctuations. On 23 May, the enhanced deviations occur to the north and west of the storm as measured by stations blmm, cnwm, slai, lthm, and nds1 from PRN 13 (Figures 2c–2d) and by stations blmm, cnwm, slai, lthm, and fbyn from PRN 23 (Figure 3). On 24 May, enhanced TEC fluctuations occurred north and west of the storm as measured by stations mbww, wylc, pltc, gdac, and amc2 from PRN 13. The deviations associated with the enhanced thunderstorm activity have more high-frequency content than typical background fluctuations. On 17 June, although the background variation is larger due to increased geomagnetic activity, it is interesting to note that TEC fluctuations with a regular period of ~4 min are superimposed on the background activity. Those fluctuations occurred south and west of the storm as measured by stations jtnt, txam, gdac, tcun, nmro, and nmsf from PRN 19. The night of 22 May 2005 has much less lightning activity than the others, and there are no clear localized regions with enhanced TEC fluctuations (Figures 2a and 2b).
 The mechanism responsible for the perturbations detected on the thunderstorm nights is not clear at this stage. Some possible causes of the fluctuations could be: AGWs produced when convective thunderstorm activity overshoots the tropopause [Alexander et al., 1995; Vadas and Liu, 2009], electrical effects from the thunderstorm itself, thunderstorm-produced density bubbles [Kuo et al., 2012], or thunderstorm-triggered Perkins instabilities [Perkins, 1973]. Based on Peter and Inan , it is unlikely that lightning-induced electron precipitation (LEP) is the cause of the perturbations in this case since they found LEP to occur ~5–8° poleward of a thunderstorm, and the fluctuations found here are localized within a few degrees of the storm. This discussion presents some individual possibilities for the cause of these localized fluctuations, but it is also possible that multiple mechanisms simultaneously affect the ionosphere above thunderstorms, leading to even more complex coupling dynamics. Future data analysis and modeling efforts are needed to understand the possible mechanisms associated with the perturbation.
 This research was supported by the Los Alamos National Laboratory's Laboratory Directed Research and Development (LDRD) project 20110184ER and 20130737ECR.
 The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.