The source of much of the observed variability within the Earth's ionosphere remains poorly understood. In this study we examine unique ionospheric measurements made above Slough, UK, for the duration of the Second World War (WWII) in order to determine whether any of the observed variability could be attributed to the major bombing campaigns across Europe.

Production of ionisation in the Earth's upper atmosphere is predominantly through photo-ionisation by solar extreme ultra violet (EUV) and X-ray radiation, while loss rates are very sensitive to the temperature and composition of the neutral thermosphere. As a result, the long-term average behaviour of the ionosphere is closely tied to solar activity and is well understood. Transient space-weather phenomena such as coronal mass ejections, high-speed solar wind streams and energetic particle events can temporarily perturb the ambient ionospheric conditions by enhancing ionospheric production through impulsive brightening of the solar atmosphere (solar flares), enhancing ionospheric loss rates (through heating of the thermosphere, affecting neutral composition and loss rates), and through direct enhancement of ionisation by precipitation of energetic particles. While the details of such processes are still the subject of ongoing research, once again, the underlying physics is broadly understood. Despite this, there are further sources of ionospheric variability that remain unaccounted for and it has been suggested (e.g. Rishbeth and Müller-Wodarg, 2006) that the source of this variability is from the lower atmosphere. Sources such as earthquakes (e.g. Astafyeva et al., 2013, and references therein), thunderstorms (Davis and Johnston, 2005; Yu et al., 2015) and explosions (e.g. Rishbeth, 1991) have been cited as potential causes of ionospheric variability, with a variety of proposed mechanisms including pressure waves, gravity waves, infrasound and modulation of the global electric circuit.

There have been a number of case studies into the impact of terrestrial explosions on the upper atmosphere (e.g. Rishbeth, 1991), most notably the events surrounding the explosion at the Flixborough chemical plant in 1974 (Jones and Spracklen, 1974; Krasnova et al., 2003), while Pohotelov et al. (1991) considered the impact on the F2 and Es ionospheric layers during a 32-day period of bombing in the Gulf War. In this current paper we make use of historical records to identify large bombing raids over mainland Europe during WWII and, using a superposed epoch, or composite analysis (Chree, 1913), look for any consistent response in ionospheric measurements made at the Radio Research Station at Slough in the UK. Using historical records, reasonable quantitative estimates of the type and tonnage of explosives for each raid can be made, enabling the raids to be subdivided by size.

While there was much military activity throughout Europe and beyond during the Second World War, some of the largest bombing raids over Berlin and other European targets took place between 1943 and 1945. The dates of these heavy raids are listed in Table 1, along with an estimate of the combined weight of high explosives used. Where no reference to the total mass of HE bombs could be found for a particular raid, an estimate was made from the recorded number of aircraft, weighted by the mean ratio of HE to total bomb load from raids for which this information was known (0.667). These dates were selected to coincide with the latter part of the war, during which intensive bombing of London was less prevalent and the raids over mainland Europe were more intense (in terms of tonnage of high explosives dropped); the length of time spent bombing was usually much shorter; and more raids occurred during the day (Middlebrook and Everitt, 1985). The times listed in this table mostly came from Berlin air-raid records (Demps, 2004). These were augmented with timings gleaned from various eye-witness accounts (see Appendix A), in particular from the archives of 550 and 410 RAF squadrons whose records have been made available online. For aircraft records, the times of the first and last recorded bombings were used. For times taken from the times of air-raid alarms, the start and end of these warnings were used. Where no detail about the length of each raid was given, the start and end times are identical. The Berlin air-raid records were recorded in local time, while military records are most likely recorded in GMT. Where known, this time difference has been taken into account, but for some this may introduce an uncertainty of 1 h into the analysis. The start times from these raids were used as “trigger” times in a superposed epoch or Chree analysis (Chree, 1913). This type of analysis is useful in geophysics for detecting a faint signal in a noisy data sequence (e.g. Davis et al., 1997). By calculating the median response of a dataset around these trigger times, any small but repeatable signal is reinforced, while any signal not associated with the trigger events is expected to occur at random and therefore cancel out when averaged. Medians were used in order to minimise the influence of outliers in the analysis.

The ionospheric data used in this study contain a strong seasonal variation introduced by annual and solar cycle variations in ion production and loss. In order that the seasonal distribution did not dominate the signal in any superposed epoch analysis, a 30-day running median was subtracted from these data. Data from the resulting parameter, δfoF2, were then aligned according to the trigger times and combined in the superposed epoch study. Since it is not possible for the ionosphere to be influenced by a given raid prior to its occurrence, the study was further constrained to ensure that each trigger event was aligned with the first subsequent ionospheric data point within a given time window. The length of time between a raid and the subsequent ionospheric measurement, along with the minimum tonnage considered for a “large” raid, are subjective variables in this analysis. In order to address this, the analysis was repeated for a range of time windows, from 0 to ≤23 h, and the size of a bombing raid was defined by the minimum estimated total tonnage of HE used, from ≥100 to ≥1000 t in steps of 100 t.

The results of a typical superposed epoch analysis using the estimated start time of each bombing raid are shown in Fig. 1. This analysis uses thresholds of ≤10 h of the ionospheric observation for raids using ≥700 t of HE for which there are 14 events. The median ionospheric response is shown as the black line in the top panel. Here the standard error in the median is shown as a grey shaded area around the line, while the dashed and dotted lines represent the 95th and 99th percentiles obtained by repeating the analysis 10 000 times using a similar number of random trigger times drawn from the same epoch for which no major bombing raid has been identified. It can be seen that the ionosphere is significantly weaker (1.9 standard deviations from the mean and around the 99th percentile level when compared with the range calculated from random trigger events) at time = 0 (the day of the bombing raids). In order to ensure that geomagnetic disturbances did not contribute to the ionospheric response, all events for which Ap exceeded 48 (Kp>5) were not considered. This is reflected in the low median values in Ap presented in the lower panel.

In order to investigate the sensitivity of this analysis to the arbitrary thresholds applied to the data, the analysis was repeated for a range of thresholds in both the length of time between the raid and the ionospheric measurement (from 0 to 23 h) and the size of a raid as defined by the minimum amount of explosives used (from 100 to 1000 t). The significance of each of these analyses was estimated by calculating the mean and standard deviation of all response times other than at time = 0 and calculating the number of standard deviations the time = 0 measurement was from this mean. The result of these analyses is shown in Fig. 2. The results are plotted on a grid with maximum time between raid and measurement along the y axis and minimum tonnage of the raid along the x axis. The significance of the response at time = 0 for each analysis is shown as a colour contour. As the thresholds used are varied, the number of raids included in each analysis will vary. This is reproduced in a similar grid also presented in Fig. 2, with the colour axis representing the number of events used in each analysis. Two results are clear from this analysis. Firstly a significant (>2σ) ionospheric response is obtained for a broad range of trigger thresholds (3 ≤ time window ≤ 7 h & 100 ≤ minimum tonnage ≤ 800 t). Secondly, the most significant result (2.99σ) is seen for a time window of ≤5 h and a minimum tonnage per raid of 300 t. The results of this analysis are presented in Fig. 3, for which there are 20 events. Consideration of the number of events contributing to each study suggests that 14 or more events are required before the random noise is reduced to a level where a significant signal can be detected. The ionospheric responses for the analyses that contain the most events (with thresholds exceeding 15 h) are not significant, indicating that these events do not contribute to the response observed within 5 h of the ionospheric observations.

From the analysis undertaken in this study, no minimum threshold in HE tonnage is resolved. In order to investigate this, data for smaller raids need to be included. In addition to the major raids considered here, there are many more smaller-scale raids involving fewer or smaller aircraft. For example, Mosquito aircraft were used in many hundreds of bombing raids throughout this period (Middlebrook and Everitt, 1985). Given the fact that the current list of major bombing raids used in this study is by no means comprehensive, it is likely that information about many hundreds of smaller raids would need to be included in order to tease out the signal of such raids from the background ionospheric variability in which further large raids were occurring. As such we consider this beyond the scope of the current study.

For the ionosphere at the altitude of the F2 region (∼200–300 km) above the UK to respond to bombing raids conducted at distances up to 1000 km away, the bombing must have generated pressure waves that were capable of propagating to ionospheric altitudes. A sound wave travelling this distance in the lower atmosphere would arrive within an hour. The speed of sound is temperature-dependent and the temperature decreases with altitude in the troposphere and mesosphere before increasing in the thermosphere. Since the thermosphere represents the most significant fraction of the vertical profile, it is likely that a sound wave propagating vertically as well as horizontally would arrive even sooner. One potential mechanism therefore is of a pressure wave propagating upwards in all directions. At higher altitudes its amplitude increases until it breaks in the upper atmosphere, depositing its energy as heat. A very rough estimate of the anticipated thermospheric temperature rise can be obtained by considering the specific heat capacity of the atmosphere which can be expressed as Q=CpnΔT, where Q is the energy input in joules (4.184×1012 for 1000 metric tonnes of TNT), Cp is the molar-specific heat capacity of N2 (∼29.1 J mol-1 K-1), n is the number of moles of gas m-3 (at ionospheric altitudes, the number density of the atmosphere is ∼1016 m-3 or 1.66×10-8 moles m-3) and ΔT is the change in temperature (K). Assuming the energy is equally distributed throughout a cylinder of atmosphere 1000 km in radius and 300 km in height, this gives a temperature rise of ∼9 K.

The dominant ion species in the mid-latitude ionospheric F region is O+, whose loss rate is temperature-dependent (Rees, 1989). However, the dominant mechanism by which O+ ions are lost is through reaction with N2 and O2 molecules in the reactions O++N2→NO++O,O++O2→O2++O. The overall loss rate, β, for O+ ions can therefore be expressed as β=k1⋅N2+k2⋅[O2], where [N2] and [O2] are the concentrations of N2 and O2 molecules respectively and k1 and k2 are the rate coefficients for the two reactions. These rate coefficients are also temperature-dependent (Rees, 1989). The combined loss rate for O+ ions is therefore dependent on both reaction rates and the concentration of thermospheric species. Müller-Wodarg et al. (1998) modelled the ionospheric and thermospheric response to localised thermospheric cooling (≤40 K) during a total solar eclipse. They predicted an 8 % increase in foF2 (∼0.2 MHz) due to the contraction of the atmosphere and an increase in the [O] / [N2] ratio caused, in part, by a reduction in the concentration of N2. It is reasonable to assume that the atmospheric expansion due to energy from localised bombing raids would have an analogous, if opposite, effect on the ionosphere and thermosphere. A rise in the background thermospheric temperature would result in an enhanced loss rate, with the equilibrium between production and loss being established at a lower peak electron concentration. Such equilibrium would be reached within minutes of perturbation, well within the resolution of the ionospheric data. Grandin et al. (2015) studied the impact on foF2 of high-speed streams on Earth. They found that a thermospheric temperature increase of 20–50 K may result in a decrease in foF2 by 0.5–1.0 MHz.

If the bombing resulted in the generation of shock waves or atmospheric gravity waves, their horizontal propagation speed would need to be of the order of 300 km h-1, while the vertical velocity component would need to be around 100 km h-1 in order to affect the atmosphere above Slough. There is evidence that turbulence generated in the lower thermosphere by space shuttle launches can propagate 1000 km horizontally within 8 h (Kelley et al., 2009). While this example was specific to the lower thermosphere at altitudes between 100 and 115 km, it nevertheless has a similar time constant to that observed for the ionospheric response to bombing in the current study. Such a mechanism may therefore contribute to the observed effect.

Infrasonic waves generated by explosions are launched preferentially in a vertical direction (e.g. Blanc, 1985). Any impact on the upper atmosphere overhead would then require horizontal transport to move that region over Slough. For the scale sizes involved this would require winds of the order of 300 km h-1 to blow consistently in a westward direction for the impact to be observed within 3 h, as suggested by the data. For this to happen continuously throughout the epoch being studied is unlikely. Whatever the cause of the ionospheric depletion, the impact appears to last less than 24 h, since only the subsequent noon value is affected.

One metric ton of TNT has an explosive energy of 4.184×109 J, which is of the same order of energy as a cloud to ground lightning stroke. While a ground-based explosion and a lightning stroke are somewhat different in location and duration, it is not unfeasible that wave energy generated by lightning could also influence the ionosphere in a similar way. Since the occurrence of lightning has distinctive diurnal and seasonal cycles, it is feasible that this mechanism could contribute to the observed seasonal anomaly in ionospheric F-region electron concentrations (Rishbeth and Müller-Wodarg, 2006).

We have shown that the influence of major bombing raids over Europe during the latter half of WWII caused a significant depletion of the ionosphere above Slough, UK. This depletion is consistent with an enhanced ionospheric loss rate resulting from thermospheric heating via the dissipation of wave energy in the upper atmosphere. While the list of bombing raids used in this analysis is by no means complete, it is nevertheless sufficient to reduce the background noise in a composite analysis to a level where the ionospheric response to bombing is statistically significant. No lower limit to the minimum mass of explosives required to generate such a response was revealed in this study, though raids using ≥100<800 t of HE were observed to deplete the ionosphere above Slough if they took place between 3 and 7 h before the ionospheric measurement. It is possible that raids occurring outside of these thresholds could still produce an effect, but the current study contains insufficient data to test this.

One metric ton of TNT has an explosive energy of 4.184×109 J, which is of the same order of energy as a cloud to ground lightning stroke. Since the occurrence of lightning has distinctive diurnal and seasonal cycles, it is feasible that a similar mechanism could contribute to the observed seasonal anomaly in ionospheric F-region electron concentrations (Rishbeth and Müller-Wodarg, 2006).

The duration of the ionospheric effect appears to be limited to within 24 h. This is currently restricted by the resolution of the digitised data. While hourly foF2 data were digitised from the original analogue ionograms, these data could be re-examined to increase the temporal resolution and investigate the behaviour of other parameters such as the D, E, F1 and sporadic-E layers. The ionospheric sporadic-E layer has already been shown to be enhanced by terrestrial lightning (Davis and Johnson, 2005; Johnson and Davis, 2006; Davis and Lo, 2008). Examining the response of this layer to terrestrial explosions would also provide further information on mechanism(s) involved in this process. The digitisation of these records is therefore highly desirable for a range of different research topics.