The Lifetimes of Plasma Structures at High Latitudes

We present an investigation of polar cap plasma structure lifetimes. We analyze both simulated data from ionospheric models (International Reference Ionosphere model and Mass Spectrometer Incoherent Scatter model) and in situ data from the Swarm satellite mission (the 16 Hz Advanced Plasma Density dataset). We find that the theoretical prediction that E‐region conductance is a predictor of F‐region polar cap plasma structure lifetimes is indeed supported by both in situ‐based observations and by ionospheric models. In situ plasma structure lifetimes correlate well with the ratio of F‐ to E‐region conductance. We present explicit predictions of small scale (∼1 km) structure lifetimes, which range from less than 1 h during local summer to around 3 h during local winter. We highlight a large discrepancy between the observational and theoretical scale‐dependency of decay due to diffusion.

where Σ k j is the height-integrated Pedersen conductivity for the regions k = E, F, and D ⊥,j is the height-integrated perpendicular diffusion coefficient, both for species j = i, e. In reality, the Pedersen current is primarily carried by ions, and so the height-integrated Pedersen conductivity can be defined in terms of the ion conductivity only, Σ Σ k k i  .
With this simplification, it is now instructional to write Equation 1 as, , , Since the diffusion coefficient of species j is proportional to mass of that species, D ⊥,i ≫ D ⊥,e . Given that the ratio of E-region to F-region conductance is substantially greater than D ⊥,e /D ⊥,i , we can simplify further, Physically, Equation 1 illustrates that a strengthening of Pedersen conductivity in the E-region as opposed to the F-region shorts out the ambipolar electric field, causing F-region plasma to diffuse at the high ion perpendicular diffusion rate instead of the balanced ambipolar diffusion rate (the applied magnetic field causes ion rates to be much higher than the electron rates, the reverse of the situation without such a magnetic field [Moisan & Pelletier, 2012]). As shown by Equation 3, as long as the ratio of E-region to F-region conductance is not negligibly small, the electrons play no role in the perpendicular plasma diffusion considered here. The electrons are frozen in the magnetic field lines, while, as mentioned, the ions carry the Pedersen currents. Incident sunlight photo-ionization, which typically causes the E-region conductivity, and is thus the primary driver for fractional factor in Equation 3, displays a strong seasonal dependence in the polar caps. This is the primary driver for observed seasonal dependencies in polar cap plasma irregularity dynamics (Basu et al., 1988;Danskin et al., 2002;Jin et al., 2018;Kelley et al., 1982;Kivanc & Heelis, 1998;Milan et al., 1999;. Let us now turn to the subject of an observable quantity related to the perpendicular diffusion coefficient: structure lifetime. In general, the time scale associated with a diffusion process adheres to the following equation (Huba & Ossakow, 1981;Moisan & Pelletier, 2012), where  is a characteristic scale length, and D is the mentioned diffusion coefficient. Plugging Equation 3 into Equation 4, we can express an estimate for F-region polar cap perpendicular plasma structure lifetimes τ ⊥ as predicted by Vickrey and Kelley (1982), Note that small-scale high latitude F-region plasma structures are believed to be generated through instability processes and be the result of the balance between production and decay (Tsunoda, 1988). Consequently, the growth of plasma structures may also effectively increase τ in Equation 4.
Unfortunately, the subject of plasma structure lifetime is rarely explicitly addressed. However, the lifetimes of polar cap patches have been documented in the literature. Due to chemical recombination, the density of polar cap patches decay toward the ambient plasma density (Wood & Pryse, 2010). Through the application of ionospheric modeling, Schunk and Sojka (1987) likewise concluded that the lifetime of a typical polar cap patch during local summer is 4 h, while during local winter the lifetime is 11 h, with lifetime defined as a decay of a patch from 10 times the ambient plasma density to a density 10% higher than the ambient plasma. Using all-sky imager data, Hosokawa et al. (2011) calculated the decay time of polar cap patches due to chemical recombination, in a case study of an equinox patch. They concluded that the lifetime of the patch in question was highly altitude-dependent, with the shortest lifetime being 1 h, at an altitude near the F-region peak of 250 km, where lifetime was defined as a reduction to 1/e times the original patch density.
The driving force behind polar cap patch decay, chemical recombination, is not dependent on plasma structure scale, and is a competing process to the ambipolar diffusion (Equation 3). The dominant ion is O + , and charge exchange collisions of O + with neutral species, which result in 2 N  and 2 O  , are the main processes through which ions diffuse vertically (Hosokawa et al., 2011;Johnsen & Biondi, 1980). At altitudes above the F-region peak at around 250 km, ambipolar diffusion will be faster than the decay time due to chemical recombination, but the latter will impact plasma densities at the topside F-region through vertical diffusion. In the case of competing decay times τ c (chemical) and τ ⊥ (diffusion), the combined decay time should be given by, The present study is a follow-up investigation based on the findings in Ivarsen et al. (2019). In the previous study, we found through an automatic detection of breakpoints in the power spectral density analysis (PSD) of Swarm 16 Hz plasma density observations in the polar caps, that around 80% of sampled local summer polar cap plasma exhibited direct evidence of plasma structure dissipation due to plasma diffusion. The corresponding fraction for the local winter polar caps were around 20%. We found that this seasonal dependency is highly predictable, and is tightly connected to solar zenith angle, which in turn controls the amount of extreme ultraviolet photoionization due to solar radiation. In the present study, we intend to investigate high-latitude plasma structure lifetimes. By applying both state of the art ionospheric models, and by using data from in situ satellite missions, we find that the theoretical predictions put forth by Vickrey and Kelley (1982), namely that E-region conductance controls the F-region plasma structure lifetimes in the polar cap, is indeed supported by evidence.

Methodology
There are two aspects to the methodology developed in the present study. First, we make an estimate of plasma structure lifetimes in the polar caps based on in situ data from the Swarm mission. Second, we approach the perpendicular diffusion coefficient using ionospheric plasma models.

In situ Plasma Structure Lifetime Estimate
Ignoring irregularity production and chemical recombination, we can assume that a portion of plasma (e.g., a polar patch) is convecting anti-sunward through the polar cap, that it only undergoes diffusion, and that it diffuses at a constant rate. Our central assumption is then that a satellite orbiting through the F-region ionosphere plasma will, at any given point along the sun-midnight line, encounter plasma that has undergone convection with a constant velocity, and diffusion without further irregularity production.
Using high-resolution (16 Hz) in situ plasma density from the Swarm mission (Friis-Christensen et al., 2006;Knudsen et al., 2017), we can estimate small-scale plasma structuring using the observed PSD of the measured electron density. With a sampling frequency of 16 Hz, we can probe fluctuations for a range of scales down to about 1 km, assuming that the plasma velocity is much smaller than the satellite velocity. At high latitudes, Swarm orbit will be almost perpendicular to Earth's magnetic field lines, and so an orbiting satellite will sample plasma structures perpendicular to the ambient magnetic field.
We consider all polar cap passes between noon and midnight made by Swarm A between October 15, 2014 and July 1, 2019, at an altitude of approximately 460 km. We use data from Swarm A in this study, but the following analysis can be applied to data from each of the three Swarm satellites with similar results. For each overpass, we translate Swarm A travel time to the distance along a straight line connecting noon to midnight, where d is the distance traveled by the convecting plasma, v S is the orbital velocity of Swarm A, α is the angle made by the orbit with respect to the noon-midnight line, t is Swarm A time, and t 0 is the time at which Swarm A approaches the polar cap. We consider polar cap passes where α < 30°, and where the satellite is located poleward of ±82° at some point during the pass.
Next, we analyze the measured electron density n. In order to look at fluctuations irrespective of the background density, we consider the unitless relative density perturbations, where n 1m is a running median filter with a window size of 1 min. We perform a PSD on  n. Here, we use a variant of Welch's PSD method, which consists of averaging modified periodograms over a logarithmically spaced spectral range (Trbs & Heinzel, 2006;Welch, 1967). We use an overlapping bin size of 60 s, with a temporal resolution of 1 s. For each bin, we integrate the PSD over 29 logarithmically spaced intervals, from 0.015 Hz down to the Nyquist frequency at 8 Hz. The integral of the PSD, given a stationary process, corresponds to the root-mean-square (RMS), the square root of which is referred to as the standard deviation, σ.
We define the scale-dependent σ λ as, where S(ω) is the PSD, ω being the frequency, and λ is the midpoint of the scale interval Δλ. The integral is performed over the frequency interval Δλ, where, assuming that the plasma velocity is much smaller than the Swarm orbital velocity v S . In this framework, σ λ represents the strength of fluctuations in the plasma density at the scale λ. Some density fluctuation powerspectra made using Swarm 16 Hz plasma density exhibit noise in the highest frequencies (Ivarsen et al., 2019). As a precaution, we impose upon the computed RMS values the requirement that, σ λ > 6 × 10 −4 , a threshold found after extensive testing.
Following from the assumptions laid down so far, plasma containing fluctuations characterized by σ λ will, once it enters the polar cap, diffuse at a constant rate D ⊥ . The time evolution of a diffusion process on σ λ with the time scale τ S is characterized by the following differential equation (e.g., Moisan & Pelletier, 2012), which has the solution, In Equations 11 and 12, t c is the plasma convection time and σ λ (0) is the initial RMS value at the point of entry into the polar cap. Now, to convert Swarm orbital distance d along the noon-midnight line (Equation 7) to plasma convection time, we write t c = d/v c , with v c being the plasma convection velocity. In combination with Equation 7, we then have for the plasma convection time, For each Swarm A orbit between noon and midnight, we store the plasma convection time t c and the relative density fluctuations σ λ for all 29 frequency intervals.
Note that we use τ S to distinguish the structure lifetime from the theoretical decay time τ-as we expect that structure lifetime as estimated in the present study will deviate from the theoretical decay time. Nevertheless, we can apply Equation 5 to express the theoretically predicted structure lifetimes. As dictated by the Fourier analysis we apply, the density gradients are associated with a wavevector k ⊥ = 2π/λ, meaning that the scale  in Equation 5 is in fact λ divided by 2π,ˆ, and so, with the assumption that the convecting structures treated in the present section only undergo perpendicular diffusion, τ S in Equation 12 will approach τ ⊥ in Equation 15.
The precise location of the cusp is known to vary both in magnetic latitude (MLAT) and magnetic local time (MLT), based on conditions related to the ionosphere-magnetosphere connection and the orientation of the interplanetary magnetic field (IMF). On average, the cusp tends to be located at around ±75° MLAT, and slightly toward the pre-noon sector Lotko et al., 2014). Though varying, with the methodology outlined here, Swarm A will nevertheless on average orbit through the cusp, given that we analyze a large number of orbits.

Modeling the Effective Perpendicular Diffusion Coefficient
Our goal is to evaluate Equation 1. To this end, we need expressions for the field-perpendicular diffusion coefficients and the Pedersen conductivity height profiles, both of which depend on the collision frequencies between the plasma species. First, we use expressions from Moisan and Pelletier (2012) for collisional plasma interactions (d ⊥,j and σ ⊥,j ), which are given below. Second, we use values for the collision interaction terms between all charged particles associated with the ion species in the ionosphere, as presented in Schunk and Nagy (1980). Third, we use the International Reference Ionosphere model for the ionospheric ion species number densities and plasma temperatures (Bilitza & Reinisch, 2008;Bilitza et al., 2014), the Mass Spectrometer Incoherent Scatter model (MSIS) for the neutral number densities (Picone et al., 2002), and data from the International Geomagnetic Reference Field for the magnetic field strength (Thbault et al., 2015). These models are not meant to offer accurate descriptions of highly localized phenomena such as the cusp, which is sensitive to the ionosphere-magnetosphere coupling. We nevertheless find that when aggregated, the simulated data offer insight into the relationship between E-and F-region conductances.
The field-perpendicular diffusion coefficient (not height-integrated) from charged particle collisions is defined as (Moisan & Pelletier, 2012), where, d 0,j = k B T j /m j ν j , with k B the Boltzmann constant, T j the temperature, ω j = eB/m j the cyclotron frequency, and m j is particle mass, all for species j. ν j is the composite collision frequency, IVARSEN ET AL.
The ionospheric Pedersen conductivity is given by (Moisan & Pelletier, 2012), where m j and n j is the effective mass and number density for species j respectively.
Next, we need expressions for the height-integrations of Equations 16 and 19. The height-integrated perpendicular diffusion coefficient D ⊥,j is defined as , for species j, and where z signifies the altitude dependency. z 0 is the lowest altitude of the F-region, and N is the height-integrated plasma density, Furthermore, the height integrated Pedersen conductivity, or conductance, , Σ E F j , is defined as , for species j, and where k = E, F signifies the region, and σ ⊥,j (z) is the altitude dependent ionospheric Pedersen conductivity (Equation 19).

Results
We perform a superposed epoch analysis on the Swarm A polar cap passes. To distinguish between different seasons, we use a 131-day window centered on the December and June solstices, without specifying the year of the polar cap pass. During the period between October 14, 2014 and June 30, 2019, we registered a total of 3,366 passes in the northern hemisphere, and 1,698 passes in the southern hemisphere. The reason for the large number discrepancy is due to Swarm orbital dynamics: the polar orbit of Swarm A is inclined 2.6° from Earth's geographic axis. Compared to the northern hemisphere, the geomagnetic south pole is further away from the geographic south pole, leading to fewer noon-midnight passes occurring in the southern polar cap. Additionally, we make a distinction between passes occurring during southward and northward orientation of the IMF. That is, we distinguish between a polar cap-crossing average value of B z > 0 and B z < 0, B z being the z component of the IMF, where we use observations from the bow shock, time shifted by around 15 min, as provided by OMNI (King & Papitashvili, 2005). Whether B z is positive or negative is known to have an impact on polar cap plasma convection (Grocott et al., 2004), and a negative B z in particular is associated with stronger ionospheric plasma irregularities (Cowley & Lockwood, 1992;Kivanc & Heelis, 1998). We found that sorting by the value of B y had minimal impact on the results.
In Figure 2, we show the result of the superposed epoch analysis for the northern hemisphere, for local winter (panels a and c) and local summer (panels b and d), and for B z positive (panels a and b) and negative (panels c and d). The equivalent is shown in Figure 3 for the southern hemisphere. Each panel shows the superposed values of σ λ for the 29 frequency intervals considered, with distance (Equation 7) and magnetic latitude on the x-axis. In all four panels, a prominent peak exists near the cusp regions, for all frequency intervals. This peak is located between 200 and 500 km after the average location of the cusp (±75° MLAT). The reason for this increase might be related to the time it takes for precipitation events to be felt at the altitude of Swarm A (460 km), and the plasma structures might be undergoing growth rather than decay.
IVARSEN ET AL. To estimate structure lifetime as outlined in the Methodology section, we fit Equation 12 to each superposed σ λ curve. That is, we fit an exponential curve through the polar cap, starting from a point after the peak near the cusp region, extending 1,000 km into the central polar cap. Here, we assume a plasma convection velocity of 300 m/s for northward IMF, and 450 m/s for southward IMF, to account for the effect of IMF B Z orientation on polar cap flow patterns (Reiff & Burch, 1985), reasonable velocities for the central polar cap (Grant et al., 1995;Thomas et al., 2015). Although this estimate might not be valid at all times, and at different locations in the polar cap, the fairly long distance over which we fit Equation 12, 1,000 km, will serve to average out spatial and temporal variations in flow velocity. Cases where the coefficient of determination, or r 2 , of the fit is less than 0.8 are discarded, which stops virtually all the IVARSEN ET AL.

10.1029/2020JA028117
8 of 17 In Figure 4, for the northern (panels a and b) and southern (panels c and d) hemispheres, we plot the structure lifetimes τ S against the scale length λ at which the lifetime estimate was calculated. Panels (a) and (c) show passes where average values of B z were positive, while panels (b) and (d) show passes where B z on average was negative. λ is calculated based on the assumption that the plasma convection velocity is negligible compared to the velocity of Swarm A (Equation 10). Local winter structure times are shown in blue, while local summer is shown in orange. The vertical error bars are constructed from a Bootstrap routine. That is, we performed 5,000 iterations of the analysis on a resampled dataset, where we sampled the entire dataset uniformly at random, with replacement. This provides 5,000 individual estimates of each calculated quantity. The error bars then represent 90% confidence intervals for the respective quantity, and represents a statistical uncertainty.
10.1029/2020JA028117 9 of 17 We see that while the local summer structure times for the most part monotonically increase with increasing length scale λ, the local winter structure times do not. The very smallest scales additionally do not monotonically increase with increasing length scale. Also shown, in a dotted lines, is what amounts to a fit of Equation 4. Here, we fit, where τ 0 and m are fitting parameters determined using linear regression, and λ 0 is a length scale equal to unity to ensure correct dimensionality in Equation 22. The values of the exponent m are given in the four panels of Figure 4, with error intervals given by the 90% confidence interval of the fitting procedure. We IVARSEN ET AL.

10.1029/2020JA028117
10 of 17 We see that the smallest scale, which corresponds to frequencies between 6.6 and 8 Hz and has a scale length of 1 km, exhibits the largest seasonal contrast. To better understand this contrast, we construct a variable we refer to as wrapped day-of-year, D w , where D is the number of days elapsed since January 1 in the relevant year (day-of-year). We then make nine overlapping bins with a window size of 65 To make a general prediction of 1 km-structure lifetimes in the polar cap, we calculate D ⊥ (Equation 1), using the procedure outlined in the Methodology section. To gather a consistent picture of the polar caps, we evaluate Equation 1 systematically for a geographic grid consisting of evenly spaced points poleward of ±77° magnetic latitude, for 1,400 points in time distributed throughout the years of 2015 throughout 2018, roughly equivalent to the time period covered by the Swarm data used in the present study. All the resulting values of 1 + Σ F /Σ E and D ⊥ for each snapshot of the polar cap are then aggregated-a total of 154,000 (for the northern hemisphere) and 145,000 (for the southern hemisphere) simulated datapoints are aggregated this way. We find that the lowest ratio of E-to F-region conductance in this dataset is around 0.1, validating the assumptions leading to Equation 3.
To make a comparison with the structure lifetime estimates, we use Equation 4, with 1km 2 / k    , to express the theoretical decay time of 1 km-scaled structures, where D ⊥ is the model-based field-perpendicular diffusion coefficient (Equation 1).
In panels (a) (northern hemisphere) and (b) (southern hemisphere) of Figure 5, we show the result of this joint analysis: In color coded two-dimensioanl-histograms, we show the distribution of model-based D ⊥ on the right y-axis, with the corresponding decay time (Equation 24) on the left y-axis, and the value of 1 + Σ F / Σ E on the x-axis, in a log-log representation. The color scale refers to number of datapoints per pixel. In yellow circle markers (B z > 0) and green triangle markers (B z < 0), we show the in situ estimated 1 km-structure lifetimes for nine bins between December and June solstice, with the calculated value of 1 + Σ F /Σ E for each bin along the x-axis. The error bars along the x-axis are the lower and upper quartile distributions of 1 + Σ F /Σ E in each bin, while the error bars along the y-axis are 90% confidence intervals from the Bootstrap error analysis described above, performed on each D w bin individually. An inset in the lower right corner of both panels expand a tightly clustered portion indicated by a red rectangle. The in situ based τ S estimates correlate well with the corresponding model-based 1 + Σ F /Σ E -number, with a Pearson correlation coefficient of 0.99 (B z > 0) and 0.97 (B z < 0) for the northern hemisphere, and 0.86 (B z > 0) and 0.70 (B z < 0) for the southern hemisphere.

Discussion
In Figure 4, we make the following significant observation: The estimated structure lifetime τ S increases with structure scale for local summer, where a power law with exponent around one-fourth accurately describes the scale-dependency of τ S . The exponent deviates strongly from that of the theoretically predicted exponent of 2 (Equation 4). As we shall now show, this result is wholly unexpected.
Chemical recombination of 2 O  ions at the F-region peak will constitute a competing process to ambipolar diffusion by the decay of plasma density enhancements, and will impact plasma structures at Swarm altitude by vertical diffusion. The inverse of Equation 6 implies that, where we use Equation 4, again with ˆ/ 2     , to express τ ⊥ , and where we assume that the structure lifetime τ S is a combination of diffusion and chemical decay times. Equation 25 provides a possible explanation for the missing exponent m = 2 in Equation 22; chemical recombination could form a plateau, where largescale decay time due to diffusion would effectively be drowned out by the much faster scale-independent τ c . However, no plateau is clearly visible for the large scales in Figure 4, where the m ≈ 1/4-fit describe the data with sufficient accuracy for all scales. As a solution to this apparent discrepancy, recall that we are analyzing the relative density fluctuations  n, which are scaled by a 1-minute (460 km) average plasma density.
Chemical recombination being scale-independent, n will decay at the same rate as 1m n (Equation 8), and so  n should be unaffected by chemical recombination. We can thus state that we do not observe molecular diffusion in our results.
10.1029/2020JA028117 12 of 17 Nevertheless, Equation 25 opens up an opportunity, as a plot of reciprocal observed structure lifetimes 1/τ S against reciprocal scale squares 1/λ 2 should reveal both the ambipolar diffusion coefficient D ⊥ and the constant chemical recombination decay time τ c , from linear regression slope and intercept respectively. In Figure 6, we show the result of the analysis presented in Section 2.1 applied to the absolute density fluctuations n, using the local summer data, for the northern (panels a and b) and southern (panels c and d) hemispheres, and for positive (panels a and c) and negative (  dashed lines) on face value, the ambipolar diffusion coefficient is valued between 3 and 7 m 2 s −1 , and the chemical decay rate is valued between 2 and 3 h. Though the ambipolar diffusion coefficient is in excellent agreement with the values shown in Figure 5, we believe a more thorough exploration of the discrepancy between the exponents m = 1/4 in Equation 22 and m = 2 in Equation 4 should be prerequisite to interpreting the results in Figure 6.
Looking at Figures 2-4, we now make the observation that the local winter structure times for the most part do not make the data correlation criteria, and thus allude the analysis, with the exception of the larger scales in the northern hemisphere. A possible explanation for this is that plasma diffusion during local winter is significantly reduced, which can explain the reported increase in local winter plasma irregularities (Ghezelbash et al., 2014;Heppner et al., 1993;Jin et al., 2017Jin et al., , 2018Prikryl et al., 2015) In the previous study (Ivarsen et al., 2019, where we similarly analyzed 60-second segments of Swarm 16 Hz plasma density), we found that sampled local winter polar cap plasma tend not to show a signature of structure dissipation due to diffusion. In fact, only 20% of local winter plasma density segments exhibited evidence of plasma diffusion, while, conversely, 80% of local summer spectra did so. It is then not surprising that most ensemble averages of winter polar cap pass in the present study failed the stated data correlation requirement that the coefficient of determination be less than 0.8. Another factor that complicates local winter diffusion is the degree to which a conducting E-region shorts out the ambipolar electric field . In the absence of a conducting E-region, anomalous currents could exist above the F-region, which could in turn lead to the existence of anomalous, or Bohm, diffusion in the F-region (Braginskii, 1965).
In Figure 5, the model-based estimate for 1 km-decay times, τ 1km (Equation 24), agree well with the in situ-based structure lifetime estimates, τ S (Equation 12), for both hemispheres. However, the in situ-based structure lifetimes are sensitive to the choice of plasma convection velocity, with higher velocities leading to a lowering of the value of τ S . Nevertheless, the apparent dependencies visible in both τ 1km and τ S of the 1 + Σ F /Σ E -number show clear agreement: The in situ-estimated structure lifetimes τ S correlate well with the simultaneous model-based 1 + Σ F /Σ E -number. They show correlation coefficients of up to 0.99 for the northern hemisphere, and up to 0.96 for the southern hemisphere. This is a strong indicator that the model first proposed by Vickrey and Kelley (1982) is suitable, and that the ratio of F-region to E-region conductance to a large degree predicts F-region diffusion rates, and thus the occurrence of plasma irregularities in the polar caps.
The reported agreement in how both the in situ based structure time and the model-based decay time respond to the 1 + Σ F /Σ E -number is largely only valid for the smallest scales available to investigation using the Swarm 16 Hz plasma density data, after which the seasonal contrast is much less pronounced. However, Keskinen and Huba (1990) found that high-latitude plasma irregularities should transition to a fully collisional regime around scales of 2-3 km, meaning there could be multiple scale-dependent regimes in the observable plasma diffusion in the polar caps, with diffusion primarily being observed on scales smaller than a threshold. Moreover, simultaneous growth might impact structure lifetimes, a topic that was explored recently in a paper by Lamarche et al. (2020). We believe more careful attention to growth, in addition to the use of higher resolution plasma density data, is necessary to further our knowledge about ionospheric plasma structure lifetimes. In addition, the analysis presented here is sensitive to the assumed polar cap convection velocity. In future investigations of plasma structure lifetimes, special care should be taken in treating plasma convection velocity, for example, by using methods of observing plasma drift velocity (Park et al., 2015).

Conclusion
In this study, we have approached the subject of field-perpendicular plasma diffusion and field-perpendicular plasma structure lifetimes from two angles. By using almost 5 years of in situ data from Swarm A, and by applying ionospheric models, we have made several new observations regarding structure lifetimes, decay time, and their seasonal dependencies. Both the in situ data and the ionospheric models support the claim that perpendicular diffusion in the F-region polar caps is highly dependent on the relationship between Eand F-region conductances.
Our results indicate that we are able to observe the characteristics of local summer diffusion in both the northern and southern polar caps. This leads to, for the first time as far the authors are aware, a systematic prediction of small-scale structure lifetimes in the F-region polar caps. We find that for the smallest scale investigated, which corresponds to frequencies between 6.6 and 8 Hz, with a scale length of 1 km, structure lifetimes range from 1 h during local summer to around 3 h approaching local winter. Although the seasonal contrast in plasma structure time harmonizes with reported seasonal dependencies in polar cap plasma irregularities, more work is needed to estimate plasma structure times more accurately, for example, by using higher resolution plasma density data. There is a large discrepancy in the theoretical scale-dependency of decay time due to plasma diffusion and the in situ-estimated structure lifetimes, and future investigations into the matter is called for.