SHIMM: A Versatile Seeing Monitor for Astronomy

The Fried parameter can be estimated from the SHIMM WFS data using the method described by Butterley et al. (2006) for the SLOpe Detection and Ranging (SLODAR) instrument, where a theoretical model is fitted to the time–averaged auto–covariance of the centroid values for a range of spatial separations within the WFS array. The theoretical auto–covariance map can be generated via numerical integration for a given WFS geometry and sub–aperture size.

For small sub–apertures, such as those used in the prototype SHIMM instrument, the shape of the auto–covariance function will be affected by scintillation such that a correction is required, as described in section 4.3. In the first instance, the measurement of $r_{0}$ without taking into account the effects of scintillation on the auto–covariance of the centroids is described.

SHWFS images are typically recorded in packets of a few hundred frames at a frame rate of a few tens of Hz, with an exposure time of 1–2 ms. Each data set, which yields a single seeing measurement, comprises one or more sequential packets spanning a duration of a few tens of seconds. Due to the large number of frames used for a single measurement ($N\sim 300-1500$), the statistical noise will be $\sim 1/\sqrt{N}\sim 3-5\%$. The centroids are determined for each WFS spot within each frame of the dataset by using the standard centre-of-mass equation. A sub-region of pixels is defined for each spot and an intensity threshold is applied (such that everything below is set to zero) to reduce the influence of readout noise on the centroid measurements. Applying an intensity threshold can result in an inaccurate estimate of $r_{0}$. Therefore, the choice of the threshold was optimised for the prototype SHIMM to minimise this effect. This resulted in a less than 1 $\%$ overestimate of $r_{0}$ for values greater than 0.1 m.

Common spot motions due to wind shake and telescope guiding errors are removed by subtracting the mean of the centroids over all WFS spots for each frame. As a result, common tip-tilt motions induced by the atmosphere are also removed. The calculation of the theoretical auto–covariance map also assumes that common motions are fully removed (Butterley et al., 2006). The mean centroid for each sub–aperture over the duration of the data set is then subtracted, in order to remove any static or very slowly varying aberrations of the telescope and WFS optics. Under typical wind speeds, the centroids for small WFS sub–apertures are expected to average to zero within seconds.

Finally, the spatial auto-covariance of the x– and y–centroids are calculated separately as

where $C$ and $C^{\prime}$ are the centroids at sub–aperture position $[i,j]$ and $[i^{\prime},j^{\prime}]$ respectively. The spatial offsets between the sub–apertures, in units of the sub–aperture diameter, are given as $\delta i$ and $\delta j$. The auto–covariance map is created by calculating this for every possible sub–aperture separation of the WFS array.

The value of $r_{0}$ is determined by fitting the theoretical auto–covariance model to a one–dimensional slice cut through the covariance map at $\delta i$ = 0 or $\delta j$ = 0 for x– and y–centroids respectively. In principle, the fit can be made to the full auto–covariance map in two dimensions. However, the SNR of the covariance map (which is dominated by statistical noise) reduces as you move away from the centre. From numerical simulations it was found that utilising the full auto–covariance did not improve the precision of seeing measurements with the SHIMM for data packets of a few seconds duration. Figure 4 shows examples of the one–dimensional covariance maps for simulated SHIMM data, demonstrating the expected variation as $r_{0}^{-5/3}$.

The fit to the auto–covariance can be linearized by plotting the measured ($A^{m}_{\delta i,\delta j}$) versus theoretical ($A^{t}_{\delta i,\delta j}$) auto–covariances. Noting that the auto–covariance is symmetrical about $\delta i=0$, then

where $d$ is the size of the sub–aperture. The linear relationship is shown in figure 4 for simulated SHIMM data.

Significant sources of noise in measuring the centroids of the WFS spots include the shot noise of the signal and detector readout noise. Since short exposure images are used with bright sources, the effects of sky background light and dark current are negligible. Any error introduced by removing the individual mean centroid will also contribute to the noise. The noise contribution to the auto-covariance is given by

where $\epsilon$ is the noise in each centroid and subscripts $l$ and $k$ refers to the sub–aperture positions of $[i,j]$ and $[i^{\prime},j^{\prime}]$. These describe the slopes for one axis (i.e. x– or y–centroids). From equation 10, the effect of noise decreases as the number of sub–apertures of the WFS is increased, except in case of $l=k$ (i.e. the centroid variance), as illustrated in figure 5. For this reason, the central point from the auto–covariance fit is excluded so that the effect of noise on the estimate of $r_{0}$ is greatly reduced. The centroid noise level can be estimated and monitored (for data quality control) by taking the difference between the measured centroid variance and its expected value extrapolated from theoretical fit to the auto–covariance. Since the prototype SHIMM only uses 12 sub–apertures, according to equation 10, for auto–covariances where $\delta i\neq\delta j$ the value of the auto–covariance will be reduced by $-\frac{1}{n}\langle\epsilon_{l}^{2}\rangle$. Figure 5 demonstrates this for different noise levels. This means that the fit will no longer intercept at the origin but will be offset by $-\frac{1}{n}\langle\epsilon_{l}^{2}\rangle$. However, the gradient of the fit will remain unchanged, as demonstrated in the left subplot of figure 5.

The SHIMM provides a low–resolution estimate of the OTP by measuring the scintillation index for the WFS spots, together with the correlation of the scintillation intensity fluctuations between neighbouring sub–apertures. The use of scintillation intensity correlations in estimating the OTP from SHWFS data has previously been exploited by the optical turbulence profiler COupled SLodar scIDAR (CO-SLIDAR) deployed on a large ($\sim$1.5 m) telescope. CO-SLIDAR is a crossed-beams method, similar to SLODAR, in which the OTP is recovered from the cross–covariance of both the centroids and the intensities of the WFS spots for a double star target (Robert et al., 2011). Another development, known as Single COupled SLodar scIDAR (SCO-SLIDAR), exploits the auto–covariances of both the centroids and intensities for a bright, single star target, to estimate the OTP using a WFS applied to a small telescope, similar to the SHIMM (Vedrenne et al., 2007). The main difference to the SHIMM is that SCO-SLIDAR uses smaller sub–apertures so that the scintillation signal is stronger and is correlated over larger WFS separations, but a very bright target is needed and the FWHM of the WFS spots are larger and therefore will be less sensitive to the image motions due to turbulence. For the SHIMM larger sub-apertures are used, resulting in weaker scintillation. However, the larger sub–apertures permit the use of fainter targets and hence continuous monitoring.

Atmospheric turbulence comprises contributions from layers at a range of altitudes, often including a strong ground layer. High–altitude turbulence results in the overestimation of $r_{0}$. Therefore, it is not possible to distinguish whether a measured value of $r_{0}$ results from a relatively strong but high altitude layer or a weaker low altitude layer. For example, for the prototype SHIMM, a measured $r_{0}$ of 0.1 m could be due to a turbulent layer at the ground with $r_{0}$ = 0.1 m or a turbulent layer at 16 km altitude with $r_{0}$ = 0.065 m.

For SHIMM data, measuring $\sigma_{I}^{2}$ and Corr makes it possible to distinguish between the scenarios described above. Both the correlation and the scintillation index increase with propagation distance. However, the correlation additionally increases with decreasing $r_{0}$ values. The scintillation index is calculated using equation 7. The correlation between neighbouring sub–apertures is given by

where

and $I$ is the time-varying intensity of a single sub–aperture $A$ or $B$. The Corr values for the SHIMM data are calculated by averaging the correlation over all instances where $A$ and $B$ are neighbouring sub–apertures in either the x– or y–direction.

A three–layer model at chosen fixed altitudes was created, empirically through simulation, such that three unknowns (the turbulence strength at these altitudes) are estimated from three measurable quantities; $\sigma_{I}^{2}$, Corr and the pre-corrected estimate of $r_{0}$. Here the model is defined with layer altitudes of 0, 5 and 15 km, chosen so each layer produces a distinguishable response in the SHIMM measurements i.e. they have a close to orthogonal response in the fit. In more detail:

• •

  The Ground Layer (0 km): The ground layer turbulence does not cause any scintillation effects, but does contribute to the total integrated seeing and hence the magnitude of the centroid covariance. Furthermore, typically there is always significant optical turbulence at the surface level, resulting from the interaction of the wind with the ground, and the heating or cooling effect of the ground on the air above it.

• •

  The Higher Layer (15 km): Higher altitude layers produce strong scintillation and spatial intensity fluctuations on relatively large scales (a few cm) so there will be a larger value of $\sigma_{I}^{2}$ as well as a significant correlation of the intensities between neighbouring sub–apertures of the SHIMM. Typically, there is also strong turbulence at the altitude of the jet stream, in the region between 10 km and 20 km.

• •

  The Middle Layer (5 km): Turbulence at intermediate altitudes, approx. 3 km to 8 km, will produce moderate intensity fluctuations due to scintillation but without significant correlation between neighbouring sub–apertures of the SHIMM. Hence the SHIMM measurements support the inclusion of a third, intermediate, layer in the model, which is placed at 5 km.

Figure 6 illustrates the response of the SHIMM three–layer model of the OTP for a single turbulent layer at a range of altitudes, for simulated data. The response is not perfectly orthogonal. For example, for a layer placed exactly at 5 km the SHIMM model places 80$\%$ of the turbulence strength in the 5 km output layer, with the remainder split evenly between the 0 km and 15 km bins. However, the model is able to distinguish between turbulence at the ground, at mid and at high altitudes sufficiently well for a number of applications, for example, to give an accurate estimate of $\theta_{0}$ for AO, and to correct the measurement of seeing for the effects of scintillation. In addition, a low–resolution profile estimate of the OTP would be useful for site characterisation and queue scheduling and to estimate the level of photometric noise from scintillation for telescope observations.

Scintillation of the light from the target star leads to a reduction in the variance of the measured centroid motions, as well as a change in the auto–covariance shape, for the small sub–apertures of the SHWFS. This results in an overestimation of the value of $r_{0}$. The same effect is also relevant to the DIMM, which will also overestimate $r_{0}$ in the presence of scintillation. This effect is discussed for the SLODAR instrument in Goodwin et al. (2007).

The total measured integrated turbulence strength of the atmosphere, $J_{m}$, can be derived from $r_{0}$, as according to equation 1. By using the method described in the previous section, the turbulence strengths at 5 and 15 km ($J_{5\,\mathrm{km}}$ and $J_{15\,\mathrm{km}}$ respectively) can be estimated. If the turbulence strength is known at these individual layers, the change in the measured turbulence strength of that layer induced by its propagation distance ($\Delta J_{5\,\mathrm{km}}$ and $\Delta J_{15\,\mathrm{km}}$) can be estimated empirically, through simulation. The sum of $\Delta J_{5\,\mathrm{km}}$ and $\Delta J_{15\,\mathrm{km}}$ is equal to the total change of the measure turbulence strength and can be subtracted from $J_{m}$ to derive the true total turbulence and therefore the corrected $r_{0}$. Figure 7 shows the results of numerical simulations for the value of $r_{0}$ estimated before and after correcting for scintillation, for different turbulence profiles. The plots show results for example three–layer turbulent profiles, where the altitude profile is denoted on each plot. A range of integrated turbulent strengths were used ($J$ = $100\times 10^{-15}$ to 500 $\times 10^{-15}$ m^1/3) for turbulent layers at altitudes of 5 and 15 km with a fixed turbulent strength of $J$ = 50 $\times 10^{-15}$ m^1/3 at altitude 0 km. It should be noted that these profiles are intentionally high–altitude heavy for the purpose of amplifying this effect, this is not a typical profile.

By comparing the values for $r_{0}$ before and after correcting for scintillation, it can be seen that the estimated value of $r_{0}$ after correction is much closer to the input $r_{0}$ value. Whilst the correction is not always exact, it is at least an order of magnitude improvement in accuracy to the original estimation of $r_{0}$.

After correcting $r_{0}$ the turbulence strength at 0 km can be derived by subtracting the strength acquired from the altitudes at 5 and 15 km from the corrected total integrated strength. It is therefore possible to estimate a low-resolution three–layer turbulence profile and therefore $\theta_{0}$ as according to equation 3. Figure 8 shows simulated examples of how the input turbulence profiles are distributed in the three–layer profile. Though this is a low–resolution profile, the model can still accurately estimate the input $\theta_{0}$ within error. The error is given by the propagation of error from the uncertainty in the fit of the theoretical model.

Knowledge of $\tau_{0}$ is important for characterising the atmospheric conditions since it measures how fast the turbulence is evolving. As described by equation 5, $\tau_{0}$ is inversely proportional to the effective wind-blown turbulence velocity ($v_{\mathrm{eff}}$), which is defined by equation 6. In order to estimate $\tau_{0}$ the SHIMM required a camera with a faster frame rate. The prototype SHIMM was therefore upgraded to an 11 inch aperture telescope and utilised a 640 × 480 Mono Prosillica GE 680 camera. This increased the number of sub–apertures used from 12 to 20, and sub–aperture size from 4.1 cm to 4.7 cm. The description of the method for estimating $\tau_{0}$ presented here is for this updated prototype SHIMM.

The effective wind velocity can be estimated by acquiring the power spectrum of the Zernike defocus mode of the wavefront aberration. The defocus term has been used previously by the FAst DEfocus monitor (FADE) instrument to estimate $\tau_{0}$ by employing a small telescope with a central obstruction to produce a ring-like defocused image (Tokovinin et al., 2008). Here, however, the Zernike analysis is applied to the WFS centroid data.

The first order Zernike modes of the atmospheric aberration, representing the angle of arrival fluctuations of the starlight or tip/tilt of the wavefront, have the largest variance. Hence, in principle, the power spectrum of the first order modes would provide the highest SNR for estimation of the $\tau_{0}$. However, the tip/tilt modes include the effects of telescope shake and guiding errors, which cannot be distinguished from the atmospheric contribution. Therefore, $\tau_{0}$ is estimated from the power spectrum of the second order defocus atmospheric term, which is unaffected by telescope shake and guiding errors. Furthermore, the defocus mode has circular symmetry so that its shape is independent of the wind direction relative to the WFS, this was verified in simulation. The SHWFS slopes are converted into Zernike coefficients via an interaction matrix describing the local gradient of each Zernike mode across each subaperture, which can be analytically computed from gamma matrices (Noll, 1976; Townson et al., 2017). The dot product of the pseudo-inverse of this matrix with measured slopes then provides the Zernike decomposition for a particular frame.

Utilising a linear-log scale, to display a normalised power spectrum, emphasizes the sharp localization of the energy at peak frequency ($f_{\mathrm{peak}}$) (Hogge & Butts, 1976). This is defined as

where $\Phi\left(f\right)$ is the power spectrum of the defocus term. In addition, the area under the curve is equal to the total energy of the power spectrum (Roddier et al., 1993). The parameter $f_{\mathrm{peak}}$ is related to the velocity by

where $\gamma$ is a constant factor related to the WFS geometry.

In practise, the measured power spectrum will comprise contributions from multiple turbulent layers of the atmosphere, each characterized by its wind speed and optical turbulence strength. One way of determining $v_{\mathrm{eff}}$ is by fitting a linear sum of several power spectra to the measured spectrum. However, a different method was developed that proved, in simulation, to be more robust for profiles with many non-distinct turbulent layers.

The method adopted here is illustrated in figure 9. First, the data is smoothed (depicted by the red line) by using a moving average of the power spectrum to reduce the scatter. Since there are fewer samples at low frequencies the moving average is completed in two parts for frequencies above and below log(0.5), where each is averaged over a different number of samples. An interpolation was used to obtain points at equal intervals (for example those denoted by the black arrow markers in the left subplot of figure 9). Each marker assumes a power spectrum of particular strength and speed. Summing these individual spectra forms a power spectrum similar to that of the original power spectrum. A value for $v_{\mathrm{eff}}$ is then found via the weighted sum of these contributions, using equation 6.

Optical turbulence profiles with more than one layer will result in the superposition of multiple peaks, each corresponding to a different layer. Figure 10 (left) shows a simulated example of this, for a two layer profile with distinct turbulent layers. Figure 10 (right) shows the results when applying the same method described above but with two layer turbulence profiles. The error in these values is due to the uncertainty in the interpolation given by a weighted sum of the squared residuals of the spline approximation. The simulations indicate that this method works for velocity profiles with wind speeds less than 25 m s^-1. For the 200 Hz frame rate of the Prosillica camera and 11 inch aperture telescope, only wind speeds up to 25 m s^-1 can be estimated. For higher wind speeds the $f_{\mathrm{peak}}$ can no longer be sampled, and only the left side (low frequency) gradient of the peak is observed. In order to measure faster layers either a detector with a faster frame rate is required or a larger telescope aperture. However, results from Stereo–SCIDAR, at La Palma in June and October 2015, show that turbulent layers with wind speeds greater than 25 m s^-1 typically occur $\sim$17% of the time.

The results presented here were from observations that took place on the 27th - 29th June 2015 and 5th October 2015 on the roof of the Isaac Newton Telescope (INT) building, La Palma, with the prototype SHIMM.

Figure 11 shows the measured profile sequences for each night of observation, demonstrating how the estimated OTP evolves with the measured scintillation parameters ($\sigma_{I}^{2}$ and Corr). The turbulence profiles have been corrected for airmass. Intervals in the profile sequences were due to target changes, telescope autoguiding corrections and occasional overcast.

As expected, larger values of $\sigma_{I}^{2}$ are reflected by increased turbulence strength in one or both of the 5 km and 15 km layers. When the correlation of intensity fluctuations between neighbouring sub–apertures is relatively low, e.g. for most of the night of 27th June 2015 and 29th June 2015, the estimated strength of the 5 km is larger than for the 15 km layer. This situation is reversed on the nights of 28th June 2015 and 5th October 2015, when the correlation is much higher.

Figure 11 also includes concurrent Stereo–SCIDAR profiles that have been binned in altitude to match the vertical profile resolution of the SHIMM. The Stereo–SCIDAR profiles broadly match those from the SHIMM and show similar trends. For example, figures 11 (a) and (c) show a relatively weak high altitude layer measured by both instruments, whereas figure 11 (d) shows a relatively weak intermediate altitude layer.

Figure 12 compares the turbulence strength at 0, 5 and 15 km (after airmass correction) between the SHIMM and Stereo–SCIDAR. The estimated strength of the ground–level turbulence is clearly correlated between the two instruments, but the SHIMM systematically measures stronger turbulence than Stereo–SCIDAR for this altitude. Figures 12 (b) and (c) show closer agreement for the higher altitude (5 km and 15 km) layers, but with substantial scatter, and some bias to overestimate the turbulence at the 15 km layer.

Figure 12 (d) compares the overall seeing angle values for the SHIMM and Stereo–SCIDAR. In most cases, the SHIMM seeing value is higher than for Stereo–SCIDAR, largely as a result of the excess turbulence strength measured by the SHIMM at ground level. The instrument response function (figure 6) shows that a small percentage of higher altitude turbulence may be incorrectly allocated to the ground layer, leading to an overestimation of this layer. However, since the turbulence at higher altitudes is typically weaker, this effect is likely to be small. Also if this was significant it would result in an under–correction of the seeing. It is likely that this measured excess results from local turbulence at the site of the SHIMM that did not affect the Stereo–SCIDAR measurements. The SHIMM was mounted on the roof of the INT building. Local turbulence may have been generated by heating and local wind–shear effects caused by the building itself, or turbulence within the closed telescope tube.

There are a number of instances where the SHIMM estimates extremely weak turbulence in one of the higher altitude layers. These appear as groups of outlying points in figures 12 (b) and (c) and are marked in green and red respectively for reference. We believe that these result from two effects in the model fit to SHIMM data, as follows.

When the turbulence at high altitudes is strong, both $\sigma_{I}^{2}$ and Corr are large. In this scenario, the SHIMM model correctly allocates strong turbulence to the 15 km layer. However, the value of the turbulence strength in the intermediate 5 km layer can become poorly constrained in these circumstances, i.e. its value no longer has a large effect on the scintillation parameters. The distribution of turbulence strength between the 0 km and 5 km layers is then more poorly constrained and noisy. In many cases, the SHIMM model fit chooses the minimum allowed value for the 5 km turbulence strength, producing the cluster of points shown in green in figure 12 (b). This effect can also be seen in figures 11 (b) and (d) for the nights of 28th June 2015 and 5th October 2015, when the 15 km turbulence is consistently strong. The SHIMM turbulence strength for the 5 km layer is very variable and is not matched by similar variability of the Stereo–SCIDAR measurements. Although this is a limitation of the SHIMM method for profile estimation, we note that in these circumstances and according to equation 4, the high altitude (15 km layer) will dominate the estimate of $\theta_{0}$ from SHIMM as well as the correction of r₀ for scintillation effects. The increased uncertainty of the 5 km layer strength is then relatively unimportant.

Conversely, when the turbulence at high altitudes is very weak, both the scintillation index and correlation values become small. In these circumstances, the turbulence strength in the 15 km layer was often underestimated by the prototype SHIMM, in particular on the night of 27th June 2015. In a number of cases, the SHIMM analysis chooses the minimum allowed value for the 15 km turbulence strength, producing the cluster of outlying points shown in red in figure 12 (c). We believe that this most likely results from the increased sensitivity of the technique to the effects of shot noise and, in particular, the readout noise of the detector when the scintillation is very weak. For example, higher readout noise results in some underestimation of Corr, and hence the strength of the 15 km layer in the model fit. We expect that this issue will be greatly reduced for more recent camera models, in particular new CMOS cameras capable of delivering high frame rates with very low readout noise. This error in the estimation of the high layer in weak scintillation has minimal impact on the correction of r₀ for scintillation effects. It is more significant for the estimation of $\theta_{0}$, resulting in an overestimation of its value, as indicated in figure in 13.

Figure 13 compares the estimated $\theta_{0}$ from the low-resolution profile of the SHIMM to that estimated from the high-resolution profile of Stereo–SCIDAR. Figure 13 additionally shows instances where the SHIMM underestimates $\theta_{0}$, due to the overestimation of the turbulence strength at the high-altitude layer. This slight bias may be due to the response of the model used, as shown in figure 6. Stereo–SCIDAR is a much higher resolution instrument and therefore will determine the strength of individual turbulent layers. The SHIMM however, will split the placement of the turbulence of the layers between the mid and high-altitude.

Observations were made simultaneously with two identical prototype SHIMM instruments located side-by-side on the roof of the INT building, between 25th - 30th June 2015 and 30th September - 5th October 2015. The results are shown in figure 14, indicating very good agreement between the two SHIMMs when they observe the same target. Some scatter in the values is expected due to statistical noise, since they observe along slightly different lines of sight to the target star. In particular, the effects of any local turbulence would be slightly different for the two instruments and would be slow to converge in the measurements. However, there is no significant bias in the measurements over a large range of seeing conditions, indicating very good repeatability in the construction and calibrations of the instrument.

The arrangement of two identical SHIMMs (SHIMM1 and SHIMM2) operating simultaneously also permitted direct testing of the limiting target magnitude of the instrument. SHIMM1 observed a bright reference star Vega (V = 0.03), whereas SHIMM2 observed fainter stars at similar zenith angles and positions as the bright reference star. Figure 15 shows the results of the seeing and the noise estimated by the two SHIMM instruments when observing targets of different brightness. It can be seen that there is a good correlation between the measurements for target magnitudes V $<$ 3. The overall noise for SHIMM2 diverges significantly from that of SHIMM1 for targets fainter than V = 2.65. This indicates that the noise for fainter targets is dominated by shot noise, rather than statistical noise. Hence we conclude that for this SHIMM configuration, although third magnitude target stars can be employed, the preference is to use second magnitude stars or brighter whenever possible.

Observations were made in July and September 2016 on the roof of the William Herschel Telescope (WHT) building in La Palma, with the upgraded SHIMM. For these dates there were no concurrent Stereo–SCIDAR observations, therefore the statistics between the two instruments are compared instead.

Figure 16 shows example power spectra of the defocus Zernike term measured with the modified SHIMM. Figure 17 illustrates the normalised frequencies of $v_{\mathrm{eff}}$ and $\tau_{0}$ estimated by the SHIMM on La Palma in 2016, as well as those for the Stereo–SCIDAR over all its observations during July and October 2015. The median values of $v_{\mathrm{eff}}$ for the SHIMM and Stereo–SCIDAR are 10.21 $\pm$ 0.16 m s^-1 and 10.44 $\pm$ 0.07 m s^-1 respectively, with standard error. This shows good agreement in the statistical distribution of $v_{\mathrm{eff}}$, indicating that the method described in section 4.4 can be used to estimate this parameter accurately. The median value of $\tau_{0}$ for the SHIMM and Stereo–SCIDAR is 3.88 $\pm$ 0.09 ms and 5.62 $\pm$ 0.08 ms respectively, indicating that SHIMM underestimates the coherence time. Since $v_{\mathrm{eff}}$ is in good agreement, according to equation 5, this arises because the SHIMM underestimates the value of $r_{0}$ relative to Stereo–SCIDAR. This is likely a result of local ground–level seeing effects at the site of the SHIMM.