Empirical evidence for multidecadal scale global atmospheric electric circuit modulation by the El Niño-Southern Oscillation

The El Niño-Southern Oscillation (ENSO) modifies precipitation patterns across the planet. Charge separation in disturbed weather and thunderstorms drives the global atmospheric electric circuit (GEC), hence ENSO-induced precipitation changes are anticipated to affect the global circuit. By analysing historical atmospheric electricity data using a new data processing procedure based on the Carnegie curve, signals correlated with ENSO sea surface temperature (SST) anomalies are revealed. These demonstrate a persistent ENSO-GEC relationship for the majority of the twentieth century, in potential gradient data from Lerwick, Shetland and Watheroo, W. Australia. The recovered data is weighted towards the first half of the UTC day, giving a GEC sensitivity up to ∼5% °C−1 of SST anomaly in the Niño 3.4 and 4 regions of the Pacific Ocean. Transferring ENSO variability by electrical means represents an unexplored teleconnection, for example, through proposed GEC effects on stratiform cloud microphysics. The strong ENSO-GEC relationship also provides a quality test for historical atmospheric electricity data, and encourages their use in reducing SST reconstruction uncertainties.


Introduction
The global atmospheric electric circuit (GEC) is embedded in the atmospheric system, balancing charge separation in disturbed weather regions with a return current in fair weather regions (Wilson 1929). Through the GEC, positive charge is transported to upper conductive parts of the atmosphere generating a global equipotential-the ionospheric potentialwith current flow to the surface, also sustaining the fair weather electric field ( figure 1(a)). The GEC's well-known diurnal cycle-the Carnegie curve-was originally found to be correlated with thunderstorm area (Whipple and Scrase 1936), figure 1(b). As the GEC is driven by convective weather, principally thunderstorms and shower clouds, internal climate system variability is expected to yield associated variability in the GEC (Harrison 2004), with effects identified on several timescales (Lavigne et al 2017). This also represents an electrical teleconnection, as current flow through stratiform clouds leads to weak droplet charging (Baumgaertner et al 2014), with possible charge-induced microphysical effects (Tinsley 2008).
One major mode of internal climate variability is the El Niño-Southern Oscillation (ENSO), conventionally monitored through sea surface temperature (SST) anomalies in specific regions of the Pacific ocean, and with which changes in lightning have been associated (e.g. Sátori and Zieger 1999). In the first reported observations of ENSO modulation of the GEC, Harrison et al (2011) showed that the annual variation of the potential gradient (PG) recorded in December at Lerwick, Shetland, which results from the GEC current, was positively correlated with the Niño 3.4 region SST anomalies between 1968 and 1983. The positive relationship was statistically robust from 09 UTC to 16 UTC, with sampling weighted to the morning hours. Recent detailed GEC modelling has now explained and confirmed the Lerwick observations as due to ENSO modulation of current generators (Slyunyaev et al 2021a). Their positive SST relationship from 09 UTC to 15 UTC is consistent  , 1968, -1983 Niño and La Niña years dashed pink and light blue respectively, with associated mean values red and blue). (RHS vertical axis shows estimated upper atmosphere ('ionospheric') potential, VI). (d) Spatial correlation between annual December Lerwick PG 1968-1983 andSST anomalies (from ERSSTv5) in the same month. Stippling marks statistically significant correlations at the 95% level. with the Lerwick findings, but, outside this interval, the Lerwick sampling is insufficient to assess the modelled negative relationship from 18 UTC to 23 UTC. (See also SI section B).
Investigation of a continued ENSO effect on the GEC is pursued here using recently recovered PG datasets from the first half of the twentieth century, with, additionally, a new data processing method. The SST anomalies considered, e.g. from the Hadley Centre reconstruction, HadISST1 (Rayner et al 2003), or Extended Reconstructed Sea Surface Temperature (ERSST) (Huang et al 2017), are principally from the Pacific Ocean Niño 3.4 region, which covers the dateline to the South American coast (5 • N-5 • S, 170 • W-120 • W), but also for the Niño 4 region (5 • N-5 • S, 160 • E-150 • W) for the central equatorial Pacific, to allow for different regional expressions of ENSO (Kao and Yu 2009).

ENSO effects in Lerwick PG data
The Carnegie curve ( figure 1(b)) has a single maximum around 19 UTC and single minimum around 04 UTC (Harrison 2013). Figure 1(c) demonstrates that the Lerwick Observatory PG data shows, on average, the same characteristic daily variation, and the GEC influence in the Lerwick PG data is also apparent from a close correlation with the ionospheric potential V I (Harrison and Bennett 2007). Figure 1(c) shows hourly mean Lerwick December PG values for each year, with El Niño and La Niña phases separated, according to the Niño 3.4 SST anomaly being positive or negative respectively. From Harrison and Bennett (2007), the equivalent median ionospheric potential V I changes from 188 kV (La Niña) to 214 kV (El Niño), i.e. an ENSO-induced V I variation of ∼20 kV.
Conventionally, the Carnegie curve is described by harmonic analysis, through amplitude and phase coefficients in where F is the PG at time t (for t in hours UTC). A 0 represents the mean PG, and A 1 to A 4 amplitudes of the 24 h, 12 h, 8 h and 6 h harmonic contributions respectively, with ϕ 1 to ϕ 4 their associated phase angles (in degrees). Table 1 provides derived harmonic coefficients for the (relative) amplitude variations in the Lerwick PG in figure 1(c), and the standard Carnegie curve. Changes are especially apparent in the 12 h terms (A 2 and ϕ 2 ), likely to follow from contributions 12 h out of phase with the usual maximum at 19 UTC, associated with disturbed weather in Asia. Further insight into the regional sources of GEC changes has been obtained by correlating the annual December PG at Lerwick (1968Lerwick ( -1983  As PG measurements were made globally throughout the twentieth century and indeed earlier, ENSO variations in other historical PG datasets are now investigated.

Global circuit data from the southern hemisphere
Although GEC observations can, in principle, be obtained anywhere, ocean measurements or pristine continental environments are preferred (Burns et al 2017), as local effects on PG can otherwise dominate. Data processing strategies have been developed to reduce local influences and extract global circuit signals (e.g. Nicoll et al 2019). Of these, identifying common variations across multiple sites, and seeking data under fair weather conditions (Harrison and Nicoll 2018), can prove effective, either separately or in combination. Archive data sources may therefore have to be carefully processed to remove local influences, and recover underlying global circuit effects.
A particularly notable agreement between three distant atmospheric electricity monitoring sites was reported by Wait and Mauchly (1937). Consistent variations in annual PG data at Ebro (Spain), Huancayo (Peru) and Watheroo (Australia) were demonstrated, for averages across June-July-August-September (JJAS) between 1924 and 1935, using data originally classified as least disturbed.
Watheroo PG data from 1924 to 1934 are given in Carnegie Institution reports (e.g. Wait and Torreson 1948), from which the calculated monthly averages have been transcribed (Harrison 2022). These were obtained on days originally selected for fair weather, to eliminate local effects such as brush burning by kangaroo hunters (Wait 1943). Figure 2(a) shows annual PG values derived for JJAS, with the Niño 3.4 SST overplotted. Plotting Niño 3.4 SST against the average PG derived for these months shows a significant positive correlation, confirming a similar effect in the southern hemisphere winter to that observed in the northern hemisphere winter. The seasonality of the GEC itself is poorly known, limiting estimates of how the ENSO response might vary across the year.
The presence of the ENSO modulation in the southern hemisphere during the first half of the twentieth century encourages further investigation of the effect in northern hemisphere data in the same period.

Extension of the northern hemisphere global circuit data
Atmospheric electricity measurements began at Lerwick in 1924, with hourly values available from 1927 (Harrison and Riddick 2022). This suggests that a longer period of comparison with ENSO is possible than 1968-1983, previously considered in Harrison et al (2011). The measurement apparatus used at Lerwick for the first half of the twentieth century was similar to the second half, with a mechanical recording electrometer connected to a radioactive probe sensor. This earlier Lerwick data cannot, however, be used directly between 1927 and 1967 for two reasons. Firstly, local radioactive contamination from weapons tests unfortunately compromised the measurement environment after about 1954, up to the mid-1960s (Hamilton and Paren 1967). Secondly, the pre-contamination 1927-1954 measurements applied a different data selection process, using the overall character of each day's measurements. The 1968-1983 measurements previously analysed were instead selected meteorologically, for fair weather conditions. The modern fair weather method is not easily applied retrospectively, hence a new approach to remove local influences in the earlier data has been developed, and is now described.

Earlier Lerwick PG data
Lerwick PG data from 1927-1954 have been recovered and digitised, for each December following Harrison et al (2011). The values entered were restricted to least disturbed days, originally classified as '0a, 1a or 2a' . This identified days on which the PG-usually positive in fair weather-was only negative for fewer than three hours, with less than 1000 Vm −1 diurnal range. Figure 3(a) shows the 1927-1954 December data for these days, plotted against UTC hour. Wide scatter is evident. For the 1927-1954 data, the interquartile range, IQR is 79 Vm −1 , compared with the 1968-1983 data (figure 1(c)), with an IQR of 50 Vm −1 . This is because the early 'daily character' method of data classification would not have rejected electrically variable meteorological conditions, such as large PG values associated with fog. Nevertheless, the 1927-1954 values commonly lie around 100 Vm −1 and contain an underlying diurnal variation reminiscent of the Carnegie curve. Since the Carnegie curve varies by only about ± 20% around its mean daily value, values in figure 3(a) which lie well beyond this central range are unlikely to result from the GEC.

Data selection
A method for removing outlier PG values is now investigated, based on assuming that the standard Carnegie curve occurs every day and that observed PG values most likely to be associated with the global circuit will lie close to the standard Carnegie variation. Original data from the Carnegie itself has previously been reanalysed to determine the statistical range on the Carnegie curve (Harrison 2013). Hourly PG values from Lerwick falling outside this typical range on the Carnegie curve are accordingly considered outliers. To preserve the largest number of Lerwick PG values potentially carrying a global circuit signal, values within three standard errors (σ) of each hourly mean value of the standard Carnegie curve are retained for further processing, broadly equivalent to selecting ∼99% of values possibly influenced by the global circuit. Figure 3(b) shows this range on the standard Carnegie curve for November-December-January, as a relative variation.
To apply the figure 3(b) selection envelope to the figure 3(a) data, a representative central value for scaling the relative variation is needed. Figure 3(c) shows a histogram of the PG values from figure 3(a), which has a long positive tail. Due to the asymmetry and tail of the distribution, the mean and median have larger values (133 Vm −1 and 120 Vm −1 respectively) than the mode (97 Vm −1 ), hence the mode is chosen as the measure of central tendency. Figure 3(c) includes the revised distribution after applying the Carnegie curve selection envelope. This shows the removal of outliers beyond the specified range around the mode: the total number of data points is accordingly reduced, from 5664 to 1255 (22%), and the IQR becomes 22 Vm −1 . Hourly values remaining after the data selection are plotted in figure 3(d), coloured by La Niña and El Niño circumstances: qualitatively, for 12-18 UTC, the mode of the El Niño values appears generally greater than for La Niña.
To assess the 'Carnegie curve filter' further, figure 4 compares median PG values for each December, before (figure 4(a)) and after ( figure 4(b)) filtering. The range in each year is clearly reduced by the Carnegie filtering, with the number of values in each year's median falling by a factor of about five. For this data, the values retained tend towards early morning, 00-06 UTC (see also figure 3(d)).

Discussion
Informed by the Watheroo PG result, the earlier Lerwick dataset from a similar era is compared with the Niño 3.4 SST in figures 4(c) and (d). For the original unfiltered data, there is little agreement (Spearman correlation coefficient r = − 0.02). In the filtered data, however, there is much better agreement (r = 0.61), including for the strong Pacific 1940-2 event (Brönnimann et al 2004). Evidently, the Carnegie filtering retains values more strongly associated with ENSO changes and rejects the much larger proportion of values probably affected by local factors. The original choice of 3σ for the data selection range was statistically based, although arbitrary: in fact, the correlation between the filtered annual PG and Niño 3.4 index can be increased further using 2.8 σ, yielding r = 0.63, retaining 21% of the original values. The specific choice is not particularly   (1924)(1925)(1926)(1927)(1928)(1929)(1930)(1931)(1932)(1933)(1934), Carnegie filtered least disturbed days), Lerwick (1968Lerwick ( -19831927-1954, Carnegie filtered least disturbed days). critical between 2.0 σ and 4.0 σ (see SI, section C). The same methodology was also applied to the Watheroo hourly data for July, during 1924-1934, yielding values principally around 09-15 UTC.
The three PG datasets from Watheroo and Lerwick (1927Lerwick ( -1954Lerwick ( , and 1968Lerwick ( -1983 have been combined and plotted against SST for both the Niño 3.4 region and the central equatorial region, Niño 4, in figure 5, using monthly PG averages across all hours. The relationship with Niño 3.4 (figure 5(a)) shows more scatter than Niño 4 ( figure 5(b)), but the sensitivity is greater with Niño 4. The agreement in both cases indicates a GEC response to ENSO in the northern hemisphere in both the first and second parts of the twentieth century, and in the southern hemisphere during the first part of the twentieth century. With either Niño 3.4 or 4 SST anomalies (representing Eastern Pacific or Central Pacific based ENSO respectively), the earlier (1927-1954) and later Overall, the sensitivity from both hemispheres is 4.7% • C −1 of SST anomaly for Niño 3.4, or 6.9% • C −1 for Niño 4. Using the ERSST version 5, the sensitivities are (4.1 ± 0.7) % • C −1 and (5.5 ± 1.1) % • C −1 respectively (see SI, figure D1). The agreement between the PG and SST anomalies adds confidence to the SST reconstructions. As a corollary, this suggests that historical atmospheric electricity datasets can provide corroboration with SST datasets, in which uncertainties remain (Huang et al 2016).

Conclusions
In summary, this analysis confirms a global circuit response to ENSO for the majority of the twentieth century. This reveals a new route by which ENSO variations are communicated throughout the atmosphere. It also provides further evidence for the operation of the global circuit as an embedded part of the climate system. From this, it follows that atmospheric electricity data can also implicitly contain climate information.
The ENSO-GEC modulation is sufficiently strong to allow SST data to be used as a comparator against which to test PG data for GEC variations. Further, as ENSO indices are uncertain for some periods, and an effective method of processing historical PG data to extract the global circuit signals has been demonstrated, atmospheric electricity data can be considered more widely for corroborating and supporting SST reconstructions.