Stratospheric Observations of Corona Current from Loon’s Superpressure Balloons
1. Introduction
Loon LLC was incubated at X1 with the mission of providing affordable access to basic connectivity around the world using superpressure balloons filled with helium to carry communications equipment. A network of balloons, each carrying the equivalent of a cellular base station, could connect to each other, and to specialized ground gateway equipment, to provide cellular coverage to users with a standard mobile phone on the ground. Much of the data presented in this paper were acquired en route to or over Peru, where Loon provided connectivity to tens of thousands of people after a 2017 flood, and Kenya, where Loon partnered with Telkom Kenya in its first commercial network deployment. The balloons were launched from sites near Winnemucca, Nevada, and Ceiba, Puerto Rico. The superpressure balloons were navigated by accessing a variety of wind directions at different altitudes by pumping air in and out of an internal ballonet; see Bellemare et al. (2020) for a description of the autonomous navigation algorithm. The project was shut down in 2021, and a large archive of technical information has been published (Alexander et al. 2021). Figure 1 shows a photo of one of Loon’s balloons inflated on the ground but with the ballonet completely deflated.
A photo of a Loon superpressure balloon inflated on the ground (note humans for scale). The internal ballonet (deflated in the photo) is filled or emptied of air in order to adjust the balloon’s altitude during flight in order to access navigationally favorable winds. An illustration of the inflated ballonet is shown in Fig. 2.
Citation: Journal of Atmospheric and Oceanic Technology 39, 7; 10.1175/JTECH-D-21-0109.1
Loon LLC partnered with local mobile network operators to provide connectivity to mobile devices using a network of stratospheric superpressure balloons, each covering an area of hundreds of square kilometers. After observing damage to flight vehicles under suspected electrical activity, Loon added a number of sensors to characterize the flight environment in order to ensure safe operation of Loon’s flight systems. See chapter 3 of Alexander et al. (2021, 154–167) for Loon’s characterization of electrically induced vehicle damage and a description of other related sensors deployed to characterize damaging electrical activity. After initial data collection, we found that the corona current sensor provided both sensitive and reliable observations of stratospheric electrical activity. We are publicly releasing corona current sensor data from 794 000 flight hours on 252 unique flights, a complement to other weather-related observations previously released by Loon (Candido 2020) and analyzed by the research community (Friedrich et al. 2017).
Corona current measurements have previously been reported in the literature (Xin and Yuan 2016), including from the ground to measure total flash rates of storms overhead (Williams et al. 1989) and using coronasondes to measure the electrical environment within a thunderstorm (Byrne et al. 1986, 1989). There have been systematic campaigns using aircraft at 15–20 km to measure the electric field and conductivity above thunderstorms in order to better understand their relationship to Wilson currents and the global electrical circuit (Blakeslee et al. 1989). Mach et al. (2009) reported peak electric fields between 1 and 16 kV m−1, notably after removal of lightning transients. They found that the contribution from lightning field changes to the Wilson current was not significant. Moreover, storms with no detectable lightning still had measurable electric fields. Finally, some storms in their dataset had opposite polarity from the standard assumption with upward electric fields.
Robertson et al. (1942) measured corona currents of ∼100 μA associated with potential gradients of ∼164 kV m−1 at elevations between 6000 and 13 500 feet. We expected smaller electrical fields in the stratosphere and designed our circuit to measure a maximum current of 10 μA.
It is our hope that this dataset will be useful for both understanding important physical phenomena, e.g., discharges in the stratosphere (Siingh et al. 2012), and for designing and operating new stratospheric vehicles.
2. Instrument description
The corona-current detection system consisted of two main parts: a hanging wire designed to hang well below the lowest point on our system and attract corona discharge and a low-current measurement circuit that was included on our storm-monitoring PCBA.
The hanging wire consisted of a single, insulated American Wire Gauge (AWG) 16 wire that hung 3.5 m below our bottom truss deck (where most of the electronics were mounted). This length placed the bottom of the wire at the lowest point of the flight vehicle and was therefore the preferred location of corona discharge. We used a transponder antenna (the TED 11-17995) to terminate the end of our hanging wire. The main reason we chose this part was to get a well-defined and small radius of curvature on the sensor endpoint (the tip of the TED antenna has a nickel-plated sphere of diameter 5/16 in. or ∼8 mm). The weight of the antenna also served to stretch out our wire and make it hang straight.
The electrical height of our vehicle, from balloon apex to the end of the hanging corona discharge wire, was approximately 15 m (but depended on several factors, including balloon orientation and amount of air in the ballonet). See Fig. 2 for the layout.
Diagram of the corona current detection system geometry relative to the flight vehicle.
Citation: Journal of Atmospheric and Oceanic Technology 39, 7; 10.1175/JTECH-D-21-0109.1
The low-current measurement circuit consisted of a sense resistor (249k Ohms), a low-pass filter RC (cutoff frequency of 16 Hz), a unity-gain op amp buffer, and a voltage divider plus level shifter to map the voltages to a range that was compatible with our microcontroller’s analog to digital converter. The resulting circuit was able to measure currents on the hanging wire in the range −10 to +10 μA. See Fig. 3.
Schematic describing the electrical circuit for the corona current detection system.
Citation: Journal of Atmospheric and Oceanic Technology 39, 7; 10.1175/JTECH-D-21-0109.1
3. Dataset description
The dataset described in this section is now publicly available (Reid 2021), and a more detailed documentation of the data is published along with the raw data. Briefly, balloon flight paths are characterized by latitude, longitude, and pressure altitude and measured by onboard GPS and pressure sensors, respectively. Corona current observations are reported in two separate datasets based on the telemetry categories “primary” and “secondary” defined below.
The corona current detector measured the corona current that flowed off the end of a long wire dangled from Loon’s payload, which was designed to be a focal point for the formation of corona discharge when the balloon is in the presence of a large, static electric field. The circuit records a maximum current amplitude of 10 μA. Our dataset consists of events above a threshold of 0.1 μA. This lower threshold was chosen to be slightly larger than the typical noise and manufacturing tolerance errors that we saw from this sensor in the absence of storms or electrical activity. Positive current flows vertically up (away from Earth’s surface).
Loon received real-time information about each balloon’s location and internal state through satellite providers while the balloon was transiting to or from a service region with ground stations. To conserve bandwidth, Loon categorized each quantity of interest (e.g., GPS coordinates, temperature and pressure sensors) as either “primary” or “secondary” based on how critical the information was for monitoring the flight. Since many quantities (e.g., a sensor error state) change infrequently, to conserve bandwidth primary fields were reported only when they changed state from the previous report. Secondary fields were sent if they changed and if there was space in the message. If more quantities have changed than there is available space in the message, the reported quantities alternated based on last reporting time. Therefore, there was no guarantee on timeliness of delivery for a particular quantity of interest, but a 30-min delay was a typical worst-case window under typical operation. Primary fields were typically received and logged approximately every minute, with secondary fields every 15–20 min. Various latencies or interruptions could temporarily increase the spacing between messages. When connected to a ground station (either directly or through a network of Loon balloons), Loon collected “high-rate telemetry,” for which new telemetry messages were received even more frequently.
Initially, the data from the corona current detector were reported as “secondary,” and the minimum, maximum, and mean current were reported for the 30 min prior to the report. Based on minor inconsistencies in corona current related quantities reported as “secondary” versus “primary,” we suspect that the time stamp associated with the secondary telemetry message may not be precisely aligned with the edge of the time window in which the minimum, maximum, and mean currents are reported, but generally they should be very close.
After an initial analysis of secondary telemetry from the corona current sensor, we designed a primary field “corona_above_threshold_count” which reported a cumulative count of the number of seconds during which the absolute value of the corona current was above the 0.1 μA threshold. This choice provided better time resolution in order to adjust navigation objectives in real time to reduce the risk of experiencing damaging electrical activity, but does not retain amplitude information. The “corona_above_threshold_count” field was only reported after 17 November 2019, so some events in the secondary dataset do not have corresponding reports in the primary dataset. The published dataset also contains several additional columns. The quantity “threshold_frac” estimates the fraction of the time elapsed since the preceding report for which the corona current amplitude was above the threshold.
Along with the corona current detector telemetry, the public dataset contains a “background” telemetry dataset with one message selected at random for each 5-min interval and flight. We have aligned the flight telemetry with two lightning-related indicators that are commercially available in real time from BCI2: Convection Diagnosis Oceanic (CDO) and Cloud-Top Height (CTH). These products are used in the airline industry; see Kessinger (2017) for more details. Note that Loon actively navigated laterally and/or vertically away from regions of electrical activity indicated by CDO ≥ 2.5 and/or CTH within 5000 ft (∼1.5 km) of the balloon’s current height, with a buffer of up to 100 km laterally around stormy regions and vertically typically reaching at least 60 000 ft (∼18.3 km). This means that the flight’s 3D position (latitude, longitude, pressure altitude) cannot be treated as independent of the value of CDO and CTH storm indicators in the neighborhood of the flight.
In this paper we also compare our corona current measurements with observations from the Geostationary Lightning Mapper (GLM). To retain the inherent spatial resolution of the GLM data (∼10 km), we work directly with GLM level-2 events rather than groups or flashes (which are generated from clusters of GLM events). See Goodman et al. (2013) for a detailed description of the level-2 data GLM data products. Because of the low temporal resolution of our telemetry data, we have aggregated the raw GLM events to 5-min intervals on a 0.1° × 0.1° grid before interpolating this “glm event density” (events per square kilometer per day) onto the background telemetry dataset. We also report the “glm optical flux” (in watts per square meter), computed on the same grid as the event density, but instead of counting events, we accumulate the reported optical flux for each event.
The public dataset contains a “background” dataset with one message selected at random for each 5-min interval and flight, along with the CDO and CTH from BCI’s real-time data products (Kessinger 2017) and GLM event density and optical flux (Goodman et al. 2013) along the flight trajectory. See dataset documentation for further details on how these fields were computed.
Finally, we note that an examination of a small set of high data rate corona current measurements (one message per second rather than the typical ∼20-min sampling period) demonstrated that the mean current was often dominated by repeated pulses due to transient streamers, and thus does not allow us to reliably estimate the current due to semistatic electric fields above storms at the available telemetry rates. Unfortunately these data were not included when constructing the final dataset. As we discuss in more detail below, we expect the majority of events captured with our sensor to occur in rapidly changing electrical fields associated with redistribution of shielding charges after a lightning discharge rather than the semistatic fields associated with Wilson current.
4. Results
Table 1 presents our estimates of the conditional probability P(C|X) that Loon measured a corona current above our 0.1-μA threshold (C) when a Loon flight coincided with an external lightning indicator X. Broadly, Loon corroborates electrical activity ∼22%–46% of the time, which is surprisingly large given that we expect Loon’s sensitivity to vary with pressure altitude, due to changes in both the breakdown voltage and typical electric field amplitude. We conclude that our chosen threshold of 0.1 μA is sufficiently high so that the vast majority of observed currents above this threshold are expected to be associated with real atmospheric electric fields. However, future stratospheric sensors that reliably detect lower amplitude currents may provide additional information about the local electrical environment. We also constrain the rate of corona current false positives by estimating P(X|C), the probability that external lightning indicator X will be observed when the corona current exceeds its threshold; we find that for 86% of those events, at least one GLM event has occurred along the path (within ∼10 km) in the corresponding 30-min window prior to the secondary telemetry report (i.e., “GLM event density” > 0 at least once in the 30-min window). Furthermore, this may be an underestimate given our use of an intermediate grid and matching GLM events to the flight path only if they mapped to the same pixel. We safely conclude that the vast majority of corona current measurements above the 0.1-μA threshold indicate real stratospheric electrical activity. Table 1 also shows good correspondence between corona current detections and CDO ≥ 2.5 within 25 km. We suspect that it is slightly less correlated [i.e., lower P(C|X) and P(X|C)] than GLM because there is a small time lag (∼5 min) when the CDO map is updated with new lightning detections. Moreover, the CDO indicator can remain above the ≥2.5 operational threshold for 30–60 min after initial lightning detection. Loon actively navigated laterally and/or vertically away from regions within up to 100 km of CDO ≥ 2.5 and/or large CloudHeight values, using current conditions as a persistence forecast of electrical activity for the next several hours. This strategy helped avoid the extremely rare, high-intensity electrical events that could damage the flight vehicle.
Estimated probability of various events in the primary and secondary datasets, where C denotes corona current above 0.1-μA threshold and X denotes an external observation expected to be correlated with stratospheric electrical activity. For the primary (secondary) column, probabilities are per 5 (30)-min interval. Conditional probabilities P(C|X) and P(X|C) are estimated using Bayes’s theorem. Note that the primary dataset covers only a subset of time included in the secondary dataset, and GLM covers only North and South America, so the absolute probabilities cannot be directly compared between columns or row subsets.
In Fig. 4 we study the relationship between the maximum corona current amplitude in a 30-min window and the accumulated optical flux along the flight path in the previous 30 min. We use a 30-min time window because the minimum and maximum corona current values were “secondary” telemetry quantities, and so may correspond to events any time within the previous 30 min. For individual corona current reports the associated distribution of GLM optical flux is very broad; nevertheless our unprecedentedly large dataset allows us to establish a positive correlation between these indicators of electrical activity strength. We find GLM optical flux
Relationship between observed corona current amplitude and maximum GLM optical flux in the 30 min prior to the corona current report (CCR) (“GLM optical flux”). (top) GLM optical flux cumulative normalized frequency for three flight data subsets: all flight data with GLM optical flux > 0 (blue), all secondary telemetry reports with a corona current report above the 0.1-μA threshold (orange), and all secondary telemetry reports for which the corona current reached or exceeded the sensor maximum of 10 μA (green). The median GLM optical flux is 5 times (30 times) larger in the latter subsets compared to the background flight data. (bottom) The mean, median, and 68% range (blue line) of the GLM optical flux in 10 bins in corona current, chosen to include equal numbers of observations in each bin. We naively fit a power-law relationship to both the mean and median and find GLM optical flux
Citation: Journal of Atmospheric and Oceanic Technology 39, 7; 10.1175/JTECH-D-21-0109.1
In Fig. 5 we show the distribution of measured corona current amplitudes for positive and negative currents. Negative currents are ∼50% more frequent and have a 3-times-larger median amplitude. This result is counterintuitive, given that the expected Wilson current is positive. However, a close examination of Fig. 3 of Blakeslee et al. (1989) shows occurrences of a rapidly changing (negative) electric field, which they associate with lightning discharges and associated shielding charge redistribution. We expect these events are responsible for the majority of our (negative) corona current events. Therefore, while our corona current detection system was not able to measure semistatic electric fields over storm clouds, it was effective at detecting lightning activity in the balloon’s vicinity.
Current distribution for positive (upward; expected Wilson current direction) and negative currents. Note that the negative current distribution tends toward larger currents, with a 3-times-larger median amplitude (−1 vs 0.34 μA). Moreover, there are 1.4 times more negative corona current events than positive in the distribution, including 1.5 times reaching the maximum current of 10 μA.
Citation: Journal of Atmospheric and Oceanic Technology 39, 7; 10.1175/JTECH-D-21-0109.1
Finally, we point out a few other features of this new dataset that may be fruitful for further investigation:
-
If we group primary reports into “events,” where the time stamp spacing is less than 15 min between reports, about half are isolated (i.e., consist of a single message) and some last more than an hour.
-
Many reports have threshold_frac > 0.8, meaning that corona current above the 0.1-μA threshold was present at least once per second over several continuous minutes.
-
While most of Loon’s flight data are well above the cloud top, ∼1.5% are within 1000 m of the estimated cloud-top height. That subset has a 2.5-times-larger fraction of seconds above the 0.1-μA threshold (“threshold_frac”) and a 6-times-larger median corona current amplitude.
5. Conclusions
In this paper, we have established that Loon’s corona current sensor detects up to approximately half of nearby electrical activity, where “nearby” corresponds to ∼10-km horizontal separation and ∼5 km above the BCI CloudHeight estimates. The observed corona current amplitude is positively correlated with the measured GLM optical flux. We hope this dataset provides researchers with a new perspective on electrical activity and storm dynamics.
For future operators of stratospheric vehicles, particularly in the tropics, we recommend both electrical hardening of the vehicle and active navigation with real-time, tailored storm data products like CDO. Furthermore, development of short-term lightning forecasts would be extremely useful for operations in the stratosphere, particularly for slow-moving vehicles like Loon balloons.
Acknowledgments.
The authors wish to thank all of our wonderful colleagues at Loon who designed, built, navigated, launched, and landed Loon balloons, as well as those that made the balloons useful to many people on the ground. Jim Olivo (BCI) and Cathy Kessinger (NCAR) helped Loon better understand and utilize the CDO and CloudHeight products in our navigation system. We would also like to thank Andy Plumer, Greg Leyh, and Rob Carver for their contributions to the design and testing of our sensors as well as our understanding of electrical activity in the stratosphere.
Data availability statement.
Telemetry data from Loon’s corona current detector along with a background dataset providing additional context for those observations (including GLM, CDO, and CloudHeight fields) are available at Zenodo via DOI with Creative Commons Attribution 4.0 International license (Reid 2021). Further details on the datasets are provided in a PDF along with the dataset.
REFERENCES
Alexander, A., and Coauthors, 2021: Loon library: Lessons from building Loon’s stratospheric communications service. X Company Rep., 437 pp., https://storage.googleapis.com/x-prod.appspot.com/files/The%20Loon%20Library.pdf.
Bellemare, M. G., and Coauthors, 2020: Autonomous navigation of stratospheric balloons using reinforcement learning. Nature, 588, 77–82, https://doi.org/10.1038/s41586-020-2939-8.
Blakeslee, R. J., H. J. Christian, and B. Vonnegut, 1989: Electrical measurements over thunderstorms. J. Geophys. Res., 94, 13 135–13 140, https://doi.org/10.1029/JD094iD11p13135.
Byrne, G. J., A. A. Few, and M. F. Stewart, 1986: The effects of atmospheric parameters on a corona probe used in measuring thunderstorm electric fields. J. Geophys. Res., 91, 9911–9920, https://doi.org/10.1029/JD091iD09p09911.
Byrne, G. J., A. A. Few, and M. F. Stewart, 1989: Electric field measurements within a severe thunderstorm anvil. J. Geophys. Res., 94, 6297–6307, https://doi.org/10.1029/JD094iD05p06297.
Candido, S., 2020: Loon stratospheric sensor data. Zenodo, accessed 1 August 2021, https://doi.org/10.5281/zenodo.3755988.
Friedrich, L. S., A. J. McDonald, G. E. Bodeker, K. E. Cooper, J. Lewis, and A. J. Paterson, 2017: A comparison of Loon balloon observations and stratospheric reanalysis products. Atmos. Chem. Phys., 17, 855–866, https://doi.org/10.5194/acp-17-855-2017.
Goodman, S. J., and Coauthors, 2013: The GOES-R Geostationary Lightning Mapper (GLM). Atmos. Res., 125, 34–49, https://doi.org/10.1016/j.atmosres.2013.01.006.
Kessinger, C., 2017: An update on the convective diagnosis oceanic algorithm. 18th Conf. on Aviation, Range, and Aerospace Meteorology, Seattle, WA, Amer. Meteor. Soc., 211, https://ams.confex.com/ams/97Annual/webprogram/Paper314031.html.
Mach, D. M., R. J. Blakeslee, M. G. Bateman, and J. C. Bailey, 2009: Electric fields, conductivity, and estimated currents from aircraft overflights of electrified clouds. J. Geophys. Res., D10204, 114, https://doi.org/10.1029/2008JD011495.
Reid, B., 2021: Loon corona current dataset. Zenodo, accessed 1 August 2021, https://doi.org/10.5281/zenodo.5144141.
Robertson, L., W. Lewis, and C. Foust, 1942: Lightning investigation at high altitudes in Colorado. Electr. Eng., 61, 201–208, https://doi.org/10.1109/EE.1942.6436259.
Siingh, D., R. Singh, A. K. Singh, S. Kumar, M. Kulkarni, and A. K. Singh, 2012: Discharges in the stratosphere and mesosphere. Space Sci. Rev., 169, 73–121, https://doi.org/10.1007/s11214-012-9906-0.
Williams, E. R., M. E. Weber, and R. E. Orville, 1989: The relationship between lightning type and convective state of thunderclouds. J. Geophys. Res., 94, 13 213–13 220, https://doi.org/10.1029/JD094iD11p13213.
Xin, E., and H. Yuan, 2016: Development of a sensor for corona current measurement under high-voltage direct-current transmission lines. Int. J. Distrib. Sens. Networks, 12, 4243, https://doi.org/10.1177/1550147716664243.
