Near-surface air temperature (NSAT) anomalies during the 20 March 2015 solar eclipse are investigated at 266 UK sites, using operational data. The high density of observing sites, together with the wide range of ambient meteorological conditions, provided an unprecedented opportunity for analysis of the spatial variability of NSAT anomalies under relatively uniform eclipse conditions. Anomalies ranged from −0.03°C to −4.23°C (median −1.02°C). The maximum (negative) anomaly lagged the maximum obscuration by 15 min on average. Cloud cover impacted strongly on NSAT anomalies, with larger anomalies in clear-sky situations (p<0.0001). Weaker, but statistically significant, correlations were found with wind speed (larger anomalies in weaker winds), proximity to coast (larger anomalies at inland sites), topography (larger anomalies in topographical low points) and land cover (larger anomalies over vegetated surfaces). In this mid-morning eclipse, the topographical influences on NSAT anomalies were apparently dominated by variations in residual nocturnal inversion strength, as suggested by significant correlations between post-sunrise temperature and NSAT anomaly at clear-sky sites (larger negative anomalies with lower post-sunrise temperatures). The largest NSAT anomaly occurred at a coastal site where flow transitioned from onshore to offshore during the eclipse, in a situation with large coastal temperature gradients associated with antecedent nocturnal cooling.
This article is part of the themed issue ‘Atmospheric effects of solar eclipses stimulated by the 2015 UK eclipse’.
Solar eclipses are of great interest to the meteorological community because of measurable changes that they induce in routinely observed meteorological variables. This interest is reflected by a large number of studies describing eclipse-related meteorological observations in the peer-reviewed literature (for a detailed review, see Aplin ). Solar eclipse effects on near-surface air temperatures (hereafter NSATs) have been studied particularly closely, perhaps because of the relatively large magnitude of temperature changes, which are usually observable with standard operational instrumentation, and because of the widespread availability of screen temperature observations. Many studies present detailed observations from a single site (e.g. [2–10]) sometimes at multiple vertical levels (e.g. [2,3,5–7,10]), while others compare observations from multiple sites (e.g. [11–13]). Improvements in data availability have allowed analyses to be constructed using data from a much larger set of sites for some of the more recent events, sometimes with large geographical spread (e.g. [14,15]). These studies demonstrate considerable variability in the magnitude of NSAT decrease, with factors such as eclipse obscuration, eclipse duration, latitude, time of day, time of year, proximity to coast, topography, land use, cloud cover, surface wetness and wind speed all potentially having some influence on the magnitude of NSAT decreases. For example, Segal et al.  reported NSAT decreases of 1.5–6.1°C in a study of the solar eclipse of 10 May 1994 over the USA. Local variability in the absence of cloud was attributed mainly to the effects of terrain and land use differences, while, at sites experiencing partial cloud cover, temperature changes were dominated by the changes in cloudiness. Hanna , in a study of the 11 August 1999 solar eclipse over the UK, reported NSAT decreases of between 1.2°C and 2.3°C, with the larger values occurring over southeast England where skies were clearer. Although the variability in the magnitude of NSAT anomalies is well documented, it is difficult to isolate (and thereby quantify) the impact of individual influencing factors when using real data, owing to the large number of possible influences.
In 2009, the Met Office introduced a new surface observing system called the meteorological monitoring system (MMS) . Integral to this system is the near real-time collection and archiving of 1 min temporal resolution data from a network of over 280 sites. The solar eclipse of 20 March 2015 provided an unprecedented opportunity for the analysis of sub-hourly changes in NSAT across a large network of sites during a deep, partial solar eclipse (i.e. obscuration more than 80% at the time of maximum eclipse). The eclipse affected the UK in the mid-morning, when temperatures would normally be increasing with the diurnal cycle. Maximum obscuration, defined as the largest percentage of the Sun’s disc obscured by the Moon at a given location, varied from 82.5% in the extreme southeast of England to over 97.0% in northwest Scotland (figure 1). The path of totality passed over sea areas to the north and northwest of the UK, reaching the Faroe Islands, approximately 400 km north of the Scottish mainland, at 0941 UTC. The time of maximum obscuration in the UK varied from 0924 UTC in southwest England to 0938 UTC in northeast Scotland. The duration of the eclipse, defined as the time between first contact and last contact (the latter is often referred to as fourth contact), varied from 131 min in northwest Scotland to 137 min in southeast England.
In this article, the mean characteristics and variability of eclipse-induced NSAT anomalies, as observed by the MMS network, are described. The geographical spread of stations in the mainland UK (50–59° N, 1 °E–7° W) is not large enough to permit investigations into the effects of obscuration or the diurnal timing of the eclipse on NSAT anomaly. Instead, advantage is taken of the high density of sites within this region, which allows investigation of the impact of ambient weather conditions (e.g. cloudiness, wind speed) and other non-meteorological variables (proximity to coastlines, local topography and land use) on NSAT anomalies under relatively uniform eclipse conditions (i.e. small variability in obscuration and time of maximum obscuration). Collectively, the results provide a more complete picture of the spatial variability in NSAT anomaly under uniform eclipse conditions than has been reported in the literature thus far. They also pave the way for a more quantitative understanding of the relative importance of different factors that may impact upon NSAT anomaly magnitude. The results should therefore be useful for verification of model predictions of near-surface temperature response to prescribed changes in solar radiation and in the forecasting of NSAT changes during future eclipse events in locations with differing topographical and land use characteristics and under differing ambient meteorological conditions.
2. Data and methods
(a) Observational instrumentation and data processing
Data are logged at 1 min intervals for all measured variables. The data are stored on site before being sent to the Met Office HQ in Exeter, UK, for long-term archiving. While screen temperature is measured at almost all sites (266 within the mainland UK), concurrent global solar irradiance, wind and cloud observations are available at only 121 sites. Screen temperature is measured using 100 Ω platinum resistance thermometers (PRTs) housed within Stevenson screens of standard Met Office design  at a height of 1.25 m above local ground level. In most cases, the screen is situated within a grass-covered enclosure. The PRT resistance is sampled at 15 s intervals and the logged values comprise a mean of four samples. Global solar irradiance is measured using Kipp and Zonen CMP 11 pyranometers. These sensors have a response time of 1.66 s and compensate for ambient temperature. Global solar irradiance is sampled every second and logged values comprise a mean of 60 samples. Wind speed and direction are measured using the Met Office Mark 6 wind system , which comprises a three-cup anemometer and vane mounted 10 m above ground level. Measurements are made at a frequency of 4 Hz, from which a running 3 s mean is calculated. Mean wind speed (1 min) and direction are then calculated from the 3 s values. Cloud variables are measured by Vaisala CL31 laser cloud base recorders. These are relatively low-power lidars, which produce backscatter profiles up to 7.6 km above ground level. The backscatter profiles are used to determine the presence or absence of cloud and the height of cloud where present, at 1 min intervals. Cloud amounts are estimated from a weighted mean of ‘cloud’ versus ‘no-cloud’ observations. The weighting value decays exponentially with increasing time since observation, with a time constant of 40 min.
(b) Calculation of temperature anomalies
Temperature anomalies are calculated in two ways. The ‘absolute’ method involves calculating the difference between a pre-eclipse maximum temperature and an eclipse period minimum temperature. Because the temperature often continued to increase for some time after first contact (probably due to the pre-existing diurnal temperature increase), the pre-eclipse maximum (tmax) was taken to be the maximum over a 2 h period ending at the time of maximum obscuration, rather than at first contact (figure 2a). The eclipse minimum (tecl_min) was taken as the minimum over the period starting at tmax and ending at last contact. The absolute anomaly (ΔTABS) was obtained by subtracting tmax from tecl_min. Individual times of first contact, maximum obscuration, last contact and the percentage obscuration at eclipse maximum were obtained for each station, using software that performs calculations using the standard equations of celestial mechanics.
The ‘linear’ method uses linear interpolation to derive a hypothetical temperature time series over the eclipse period, using observed temperatures at the time of first and last contacts (figure 2b). This time series is assumed to be representative of temperature observations in the absence of an eclipse. For each minute during the eclipse period, an instantaneous temperature difference is calculated by subtracting the linearly interpolated temperature (tlin) from the observed temperature (tobs). The temperature anomaly (ΔTLIN) is then taken to be the minimum value of (tobs−tlin). The linear method has the advantage that it takes into account the effect of diurnal temperature changes. This may be important when the eclipse anomaly is superimposed upon large diurnal temperature changes, as would be expected for mid-morning or early evening eclipses in clear-sky conditions. Furthermore, because temperature differences are calculated every minute, the linear method allows the evolution of temperature differences to be evaluated over the whole eclipse period.
3. Overview of ambient weather conditions
On the morning of 20 March 2015, a large anticyclone was centred to the west of the British Isles, near 52° N 16° W (figure 3). Much of southern and central UK lay within a region of very weak pressure gradients, near the axis of a broad ridge extending eastwards from the centre of the anticyclone. Consequently, winds were light and variable in direction over much of this region (figure 4), with variable amounts of cloud. Over northern UK, further from the axis of the ridge, pressure gradients were stronger, with light to moderate winds from the west or northwest. A weak cold front, moving south-eastwards on the northern flank of the anticyclone, affected parts of Scotland and Northern Ireland during the eclipse period. Conditions near the front were cloudier than in some areas further south, though no precipitation was observed. Locally, the passage of the cold front complicated the evolution of surface temperatures (notably at several sites in northeast Scotland, as is discussed subsequently). Away from the front, conditions were largely quiescent with no synoptic-scale changes in ambient weather conditions over the eclipse period. Although cloud amounts were spatially variable, the regions of cloud were persistent and tended to move only very slowly, as shown by sequences of satellite data (not shown). Where conditions were clear, notably over parts of south and central Wales, winds were very light and variable in direction and temperatures locally less than 0°C at 0600 UTC. This may be compared with temperatures of 4–6°C in surrounding cloudier regions (e.g. central southern England), suggesting the presence of a well-developed nocturnal inversion in at least some of the clearer areas.
4. Results and discussion
(a) Clear-sky global solar irradiance decreases
Of the 121 sites recording cloud and global solar irradiance, eight experienced entirely clear conditions over the period 0730–1200 UTC on 20 March 2015. For these sites, an estimate of global solar irradiance in the absence of an eclipse was obtained by applying a fourth-order polynomial fit to observed clear-sky global solar irradiance data at Camborne (50.22° N, 5.33° W), UK, over the period 0730–1200 UTC on 20 March 2009. A small correction was applied to account for latitude differences between Camborne and each of the eight sites (range 51.09–53.18° N). The correction was improved by an iterative process such that the final, corrected polynomial curve produced the smallest-possible difference between the observed pre- and post-eclipse global solar irradiance time series for each station. Using this method, the mean reduction in global solar irradiance was 90.74%, compared with a mean obscuration of 86.96%. The mean decrease in global solar irradiance was, therefore, 3.78% larger than the mean obscuration (range 3.42–4.54%). These results agree well with theoretical values calculated by Segal et al. , who predicted a reduction of 92.8% in top-of-atmosphere solar irradiance for an obscuration of 89%. The larger percentage reduction in global irradiance, relative to the obscuration percentage, is probably due to limb darkening; areas near the centre of the Sun’s disc appear brighter and hotter than areas near the edge, because of the difference in radiation path length through a unit height of the Sun’s photosphere in different parts of the visible disc. The path length is shortest near the centre of the disc, and so it is possible to ‘see’ deeper into the photosphere there than near the limb .
(b) Mean temperature anomalies and comparison of derivation methods
Using the absolute method, NSAT anomalies ranged from 0.00°C to −3.80°C, with a median value of −0.70°C. Values of zero are indicative of an inflection point in the temperature trace rather than a distinct minimum (as might occur in the case of a small anomaly superimposed on a large diurnal temperature increase). Anomalies were slightly larger when the linear method was used, ranging from −0.03°C to −4.23°C, with a median value of −1.02°C and a mean of −1.18°C. The larger magnitude of linear anomalies is due to the fact that this method takes into account the diurnal temperature cycle which, given the mid-morning timing of the eclipse, resulted in higher temperatures post-eclipse than pre-eclipse at most sites (averaged over all sites, the temperature at first contact was 6.01°C, compared with 7.51°C at last contact). Exceptions to this rule occurred at several sites over northeast Scotland, where temperatures fell abruptly late in the eclipse period (by 1.5–1.9°C over a 10 min period) and remained low post-eclipse, due to the passage of a weak cold front during the eclipse (the front is evident in figure 3 over northwest Scotland). This non-eclipse-related temperature decrease resulted in small anomalies using the linear method (−0.25°C to −0.60°C), but large anomalies using the absolute method (−2.4°C to −3.0°C). The smaller linear anomalies are closer to values observed at other sites in northern and central Scotland. Because the linear method takes into account the diurnal temperature cycle and appears more robust to temperature changes unrelated to the eclipse itself, linear anomalies are used in subsequent analysis.
The mean time between maximum obscuration and the maximum negative NSAT anomaly (i.e. the lag time to maximum anomaly) was 15.68 min (median 15 min), but with large variability (range −63 to 59 min; s.d. 14.21 min). The mean and median values are within the range of values previously reported in the literature; 5–20 min appears typical . Lag times were between 0 and 30 min at 85.3% of sites. However, the lag time was smaller than zero (i.e. maximum NSAT anomaly occurred before the time of maximum obscuration) at 4.9% of sites. No statistically significant differences were found between lag times at cloudy and clear-sky sites.
(c) Temporal evolution of anomalies
All-site mean differences between observed and linearly interpolated temperatures were calculated for each minute between first and last contact, in order to show the mean evolution of temperature differences over the eclipse period (figure 5). To simplify the analysis, eclipse duration was assumed to be 134 min at all sites (actual values range from 131 to 137 min) and differences were therefore calculated using observed temperatures 67 min before and after the time of maximum obscuration at each site. The 134 min approximation results in small differences in the NSAT anomalies at individual sites, when compared with anomalies calculated using interpolation between the actual times of first and last contact at each site (mostly less than 0.2°C), but with negligible difference (approx. 0.02°C) in the all-site mean anomaly. Averaged over all sites, the largest negative NSAT difference (tobs−tlin) occurred 15 min after the time of maximum obscuration. Positive differences are evident early in the period, which illustrate a limitation of the linear interpolation method. First contact occurred approximately 2 h after sunrise, therefore falling close to the time of the maximum rate of diurnal temperature increase. Last contact, being closer to local solar noon, coincided with a period characterized by smaller rates of diurnal temperature increase. Therefore, linear interpolation will tend to underestimate the rate of non-eclipse temperature increase early in the period and overestimate it late in the period. From this, it may be inferred that anomaly magnitudes are slightly underestimated. The positive difference peaks 17 min after the mean time of first contact, though a slowing in the rate of increase suggests that the decreasing solar irradiance was impacting on temperatures (relative to the normal diurnal cycle) as early as 10–15 min after the mean time of first contact. The rate of decrease reaches a well-defined maximum of 0.038°C min−1 6 min before maximum obscuration (i.e. 21 min before maximum negative NSAT anomaly). The rate of increase reaches a maximum of 0.031°C min−1 44 min after maximum obscuration (i.e. 29 min after maximum negative NSAT anomaly).
(d) Impact of cloud cover
In agreement with the results of previous studies (e.g. [14,15]), cloud cover was found to have a strong impact on NSAT anomalies (figure 6a), with larger negative anomalies in clear-sky conditions. Using Spearman’s rank analysis, a strong positive correlation was found between NSAT anomaly and the mean cloud amount between first and last contact (R=0.7435, p<0.0001). The strength of correlation appeared relatively insensitive to the period over which the mean cloud cover was calculated. For example, using cloud amounts averaged over a 1 h period centred on maximum obscuration, R=0.7251. Using cloud amounts at maximum obscuration, R=0.6756. The association between clear skies and larger negative NSAT anomalies is strikingly apparent when contours of NSAT anomaly are overlaid on visible satellite imagery (figure 7a; note that 0900 UTC imagery is used here, because the Moon’s shadow reduced the contrast in the visible imagery close to the time of maximum obscuration). The largest negative anomalies on 20 March 2015 occurred in a swathe from southwest England to northeast England, including much of Wales and central England, where skies were largely clear for the duration of the eclipse (figure 7b). A separate area of larger negative anomalies is apparent over southeast Scotland, where skies were also largely clear. The median NSAT anomaly magnitude for clear-sky sites (less than 0.5 oktas) was 2.19°C. This is almost four times larger than the median NSAT anomaly magnitude at cloudy sites (8 oktas), of 0.56°C. However, substantial variability is also evident within the clear-sky areas (figure 7a), suggesting that other factors must have played a role in determining the magnitude of anomalies at individual sites. Clear-sky NSAT anomalies ranged from −1.23°C to −4.23°C, while cloudy-sky anomalies ranged from −0.16°C to −1.07°C. A t-test between clear (less than 0.5 oktas) and cloudy (more than 7.5 oktas) NSAT anomalies indicates that the difference is significant at the 99.9% level, despite the relatively large variability within clear and cloudy classes.
For the range of obscuration observed over the mainland UK (82–97%), the effect of cloud cover clearly dominates over obscuration effects, in agreement with the results of Hanna  for the 11 August 1999 solar eclipse. The swathe of clear skies lay almost parallel to the obscuration contours, and mainly between the contours of 86% and 88% obscuration (cf. figures 1 and 7a). Consequently, the mean NSAT anomaly for sites with 86–88% obscuration was −1.90°C, while the mean anomaly for sites with 94–97% obscuration (primarily located in northern and western Scotland, where cloudy conditions prevailed) was −0.89°C. However, some small differences that may be related to obscuration are apparent when cloudy sites are considered in isolation. The mean anomaly at cloudy sites (more than or equal to 7.5 oktas) in southeast England (82–86% obscuration) was only −0.58°C, while the mean anomaly for cloudy sites in north and west Scotland (92–97% obscuration) was −0.80°C.
(e) Impact of wind speed
NSAT anomalies were found to be weakly correlated with mean wind speed, such that larger negative anomalies tended to occur with weaker mean wind speeds, where the mean wind speed was evaluated over a 1 h period centred on the time of maximum obscuration (figure 6b). Spearman’s rank analysis yielded a correlation coefficient of 0.2434, with an associated p-value of 0.0072, indicating that the correlation is significant at the 99% level. A further check using Spearman’s rank analysis revealed no statistically significant correlation between mean cloud amounts and mean wind speeds (p=0.4404), suggesting that the correlation between wind speed and NSAT anomaly was not simply the result of lighter winds tending to occur in clear-sky conditions. Higher Spearman’s rank correlation coefficients were obtained when comparing the NSAT anomaly with mean wind speed for sites with little or no cloud (R=0.3914); however the small sample size (n=25) means that this correlation was not statistically significant at the 95% level (p=0.053).
(f) Proximity to coast
Founda et al.  showed that the magnitude of temperature anomalies in the 29 March 2006 eclipse over Greece was correlated with distance from the coast, with inland stations exhibiting larger negative temperature anomalies. Coastal impacts on NSAT anomaly were investigated for the 20 March 2015 event by classifying sites as coastal or inland by analysis of ordnance survey maps. Stations were classified as coastal if they were located within 3 km of the coastline, including large river estuaries. In order to isolate any possible coastal effect from cloud effects, sites with mean cloud amount more than 2 oktas were excluded from the analysis. Because this resulted in a small sample (n=25), non-cloud-observing sites that were situated within clear-sky areas were also included. This was achieved by overlay of site locations on visible satellite imagery. Sites located within clear-sky areas at the start and end of the eclipse were included. This increased the clear-sky sample size to 49 sites, of which 18 were classified as coastal. The mean NSAT anomaly at coastal sites was −1.90°C, compared with −2.25°C for inland sites (figure 8a). Using a t-test, the difference between coastal and inland NSAT anomalies was not found to be significant at the 95% level (p=0.078). However, if the two largest temperature anomalies (those observed at coastal Aberporth (−4.23°C) and inland Sennybridge (−4.18°C) in Wales) are treated as outliers and excluded, the difference becomes significant at the 99% level (p=0.006). The −4.23°C anomaly at Aberporth was the largest negative anomaly of all 266 temperature-observing sites in the UK, and the value is more than five s.d. below the mean anomaly for other coastal sites. Sennybridge’s temperature anomaly of −4.18°C is more than three standard deviations below the mean anomaly for inland sites. Causes of the particularly large NSAT anomaly at Aberporth are discussed subsequently.
Cooling of near-surface air in eclipses has been shown to allow development of local cold air drainage flows (e.g. [12,19–21]). This suggests that NSAT anomalies may be larger in topographical low points, where cold air can accumulate (sometimes referred to as ‘cold pooling’). Topography is often highly complex with structures on many spatial scales. For this reason, two methods are devised to describe the topographical setting of the 31 inland, clear-sky sites, so that NSAT anomalies from sites in different topographical settings can be compared. The first method involves estimation of the mean elevation of land surrounding each site using digital terrain data from the National Oceanic and Atmospheric Administration (NOAA) ETOPO1 global relief model (http://www.ngdc.noaa.gov/mgg/global/global.html). Raw data are available on a grid with horizontal spacing of 1 arc minute (1.852 km). The mean elevation of the 3.43 km2 grid box within which the site is situated was extracted and compared with the site elevation. In an attempt to take into account some of the aforementioned complexity in topographical features, mean elevation was also calculated on larger grids surrounding each site, with horizontal spacing of 5.56 km and 9.26 km (i.e. 3×3 and 5×5 grids, equating to areas of 30.87 and 85.75 km2, respectively). An estimate of the topographical situation of each site was made by taking the difference between the grid mean elevation (EG) and the station elevation (ES). Two topography classes were defined as follows:
— neutral and high-point sites: (ES−EG)>0 m
— low-point sites: (ES−EG)≤−10 m.
Note that (ES−EG) differences near zero could indicate either sites located in flat terrain or sites located within hilly areas but at an elevation that happens to be close to the mean elevation of the surrounding area (e.g. sites located part-way up a valley side). For each grid box size, mean and median NSAT anomalies for neutral/high-point and low-point classes are shown in table 1. t-tests indicated significantly larger negative NSAT anomalies at low-point sites than at neutral/high-point sites, though sample sizes are small for both classes (table 1). Spearman’s rank analysis yielded a correlation coefficient of 0.4062 between negative NSAT anomaly magnitude and values of (ES−EG) on the 85.75 km2 grid, which is significant at the 95% level (p=0.023).
The second method was more subjective and involved the use of Ordnance Survey maps, considering an area of radius approximately 10 km around each site. Sites were classified as hill top, ridge top, upland plateau, hill side, valley side, valley bottom and flat, based on the nature of the surrounding topographical features and the location of the site relative to these features (table 2). Sample sizes are too small to draw comparisons between the mean NSAT anomalies for individual classes. However, comparisons may be made between valley bottom sites (n=12) and ridge top, hill top and upland plateau sites (n=11). Negative anomalies are, on average, larger in valley bottom sites (−2.47°C) than at hill top, ridge top and upland plateau sites (−1.86°C). A t-test indicates that the difference is significant at the 99% level (p=0.0073) (figure 8b).
(h) Temperature and wind observations at two sites with unusually large NSAT anomalies
Further insight into the potential importance of local flows in determining the magnitude of NSAT anomalies at individual sites may be gained by analysis of temperature, wind speed and wind direction time series from two of the three sites with the largest NSAT anomalies: Aberporth and Trawsgoed in Wales. The meteorological site at Aberporth (52.139° N, 4.570° W) is located approximately 0.6 km to the south of a north-facing coastline, approximately 1.8 km west-northwest of the town of Aberporth itself. The site lies at an elevation of 133 m. Inland, the terrain is gently undulating, with low hills reaching elevations of between 170 and 180 m. One-minute mean wind speeds at Aberporth varied between 0.5 and 2.5 knots during the eclipse period, but with little variation in the 10 min mean speeds. By contrast, wind direction variations were large. In the hour ending at first contact, winds varied between 300° and 360° (i.e. an onshore flow). From about 0850 UTC (i.e. 23 min after first contact), the wind began to back gradually, becoming westerly by 0915 UTC. Winds continued to back through much of the remainder of the eclipse period, reaching between 200° and 230° (i.e. an offshore wind) from 1000 to 1017 UTC, before veering again to northerly near the end of the eclipse period. Temperature decreases were particularly rapid between 0928 and 0942 UTC (i.e. 1–15 min after maximum obscuration), which corresponds to the period that winds started to back to the south of west and, therefore, to the onset of winds with an offshore component. This is later than the average time of fastest NSAT decrease at most sites, which was 6 min before the time of maximum obscuration (figure 5). The lag between maximum negative NSAT and maximum obscuration was 24 min, which is 9 min greater than the mean lag time for all stations. These observations suggest that at least part of the observed temperature decrease at Aberporth was due to a change in winds from onshore to offshore, in a situation where land temperatures were substantially lower than adjacent sea surface temperatures, given antecedent nocturnal cooling (as demonstrated by areas with air temperature less than 0°C over Wales at dawn; see figure 4). This, and the subsequent re-establishment of a milder onshore flow following the eclipse, could explain why the NSAT anomaly at Aberporth was over five s.d. below the mean anomaly for coastal sites. An open question is whether the observed wind direction variations were related to the eclipse itself. The veering of the wind near last contact and, in particular, the backing after first contact and up to the time of maximum obscuration are consistent with some other observations of wind direction variations during eclipses (e.g. [7,11,13,22]); however, as yet there is no consensus on the nature of wind direction changes in the literature, with other studies showing no systematic variation (e.g. ). Wind direction changes have variously been attributed to dynamic effects associated with cooling in the umbra and penumbra (e.g. ), the development of local drainage flows [12,19–21] or suppression of sea breeze circulations (e.g. ). In this case, it is not possible to determine the cause of the wind direction changes, or even to ascertain whether they were associated with the eclipse in any way. However, it is clear that the wind direction changes impacted substantially upon the magnitude of NSAT anomaly at this location. For a comprehensive analysis of surface wind observations in the UK on the 20 March 2015 eclipse, the reader is referred to Gray .
Trawsgoed is situated in the lower reaches of the Afon Ystwyth valley, approximately 12 km from its mouth at Aberystwyth. The valley is orientated generally east to west, though the orientation is closer to southeast–northwest in the immediate vicinity of Trawsgoed. The meteorological station is situated at the northwest edge of the valley bottom, just above the approximately 1 km wide floodplain at an elevation of 63 m (52.344° N, 3.947° W). The valley is enclosed on both sides by steep hills which reach elevations of 250–350 m, but which are well dissected by the valleys of tributary streams, especially to the northeast. Wind speeds were very low at Trawsgoed before, during and after the eclipse period (generally in the range 0.5–2.0 knots). In the hour preceding first contact, wind direction varied between 35° and 105°, with a mean direction of 79°, indicative flow generally orientated down the valley. The greatest rates of NSAT decrease at Trawsgoed occurred from 0934 to 0939 UTC (i.e. 5–10 min after the time of maximum obscuration), during which the temperature decreased by 0.6°C. This occurred immediately after an abrupt change in wind direction from 100° to 74° and a reduction in wind speeds to less than 1 knot. Winds were evidently too light to move the wind vane between 0935 and 1001 UTC, because the direction remained exactly constant over this period. The wind speed fell to zero between 0945 and 0952 UTC. The greatest negative NSAT anomaly of −3.43°C occurred at 0946 UTC (i.e. 17 min after the time of maximum obscuration), near the beginning of the calm period. From approximately 1000 UTC, a westerly wind developed, and wind speeds increased from the calm recorded at the time of maximum NSAT anomaly to around 1.0–1.5 knots. Temperatures were already rising by the time of the onset of westerly winds, but the rate of temperature rise increased slightly after its onset.
Reductions in wind speed during an eclipse have been reported widely (e.g. [2,5,7,9,11,12,15,19,25]) and are usually attributed to the increase in near-surface stability which reduces turbulent mixing, sometimes allowing near-surface flow to decouple from the geostrophic flow. Segal et al.  showed that NSAT decreases were more rapid following the decoupling of winds at low levels. The reduced mixing allows greater cooling at screen height, given that cooling tends to be stronger near the surface (e.g. [2,6]). In this case, the predominance of easterly winds and depressed temperatures prior to first contact suggests that the nocturnal inversion was still in place at Trawsgoed as the eclipse developed. It was not until 31 min after the time of maximum obscuration that a westerly wind developed. The westerly flow is broadly consistent with the geostrophic flow direction given the synoptic situation (figure 3), and so its development probably signifies the dissipation of the nocturnal drainage flow at Trawsgoed and re-coupling of surface winds to the geostrophic flow. A delay in the time of break-up of the nocturnal inversion was predicated for mid-morning eclipses in the numerical simulations of Segal et al.  and was found to produce larger NSAT anomalies than in eclipses at other times in the diurnal cycle. In the 20 March 2015 event, it is interesting that for clear-sky, inland sites the magnitude of NSAT anomaly is negatively correlated with the temperature at 0600 UTC (r=−0.2569) and at 0700 UTC (r=−0.3329), suggesting that larger (negative) anomalies were more generally associated with stronger residual nocturnal inversions, in agreement with the theoretical discussions of Segal et al. . This may also account for the occurrence of larger NSAT anomalies in topographical low points, because any residual nocturnal inversion could be expected to be stronger in such locations, due to cold air drainage and cold pooling effects. Trawsgoed’s sheltered valley bottom location appears to have been particularly favourable for development of a large negative NSAT anomaly, because wind speed reductions during the eclipse, given the already light winds pre-eclipse, resulted in development of totally calm conditions for a short period after maximum obscuration, during which time mixing could not have reduced the NSAT anomaly near ground level. The 0.08 knot 10 min mean wind speed that occurred at Trawsgoed in the period ending 0953 UTC was the lowest-observed value at any clear-sky site (of those with wind observations) during the eclipse period.
(i) Land cover
NSAT anomalies have been shown to depend on the land cover type. For example, Hanna  and Segal et al.  reported larger anomalies over vegetated surfaces than over bare surfaces. Conversely, for land surface temperatures, anomalies tend to be larger over bare surfaces . The impact of land cover on NSAT anomalies in the 20 March 2015 eclipse has been investigated using the European Space Agency (ESA) Climate Change Initiative land cover map for the 2008–2012 epoch (http://www.esa-landcover-cci.org/) [27,28]. Data are available on a grid with horizontal spacing of 300 m. Land cover classifications were obtained for each clear-sky site (n=49). Mean NSAT values were compared for all land cover classifications with a sample size of more than or equal to five sites, which includes grassland, rain-fed cropland and urban categories. Statistically significant differences were found between urban and grassland (p=0.005), and between cropland and grassland (p=0.036), with larger mean anomalies over grassland in both cases (table 3). Repeating the analysis for inland sites only, statistically significant differences remain between grassland and cropland (p=0.011; figure 8c), but no statistically significant difference could be found between urban and grassland (p=0.212). This suggests that much of the difference between urban and grassland anomalies, when coastal sites are included, is due to the larger percentage of coastal sites in the urban sample (61%) than in the grassland sample (7%). The results generally agree with previous studies [14,15], in that anomalies over areas with larger fractional vegetation cover (e.g. grassland) are larger than those over areas with lower fractional cover, noting that cropland is likely to have had relatively low fractional vegetation cover in this case, given the early spring timing of the eclipse. The lack of statistically significant differences between urban and grassland sites inland may be due to the small sample size of inland urban sites (n=7).
NSAT anomalies in the 20 March 2015 eclipse ranged from −0.03°C to −4.23°C, with a median value of −1.02°C. These values are well within the range previously reported in the literature. Cloud cover, wind speed, topography, proximity to coast and land cover all had a measurable and statistically significant impact on the magnitude of NSAT anomalies. Cloud cover was clearly the dominant influence (figure 8d), as indicated by a Spearman’s rank correlation coefficient of 0.7435 (p<0.0001) between NSAT anomaly and mean cloud amount. The impact of cloud far outweighed the impact of differences in obscuration, though the range of obscuration across the UK was relatively small (82–97%). Small differences in NSAT anomalies between generally cloudy southeast England (82–86% obscuration) and similarly cloudy north and west Scotland (94–97% obscuration) were however noted, and it is likely that a larger obscuration effect may have been observed had the clear-sky conditions been experienced over a wider range of obscuration values (in the event, clear skies were largely restricted to a narrow band corresponding to obscurations of 86–88%). Where skies were clear and wind speeds very light (less than 2 knots), local variations in flow sometimes impacted strongly on NSAT anomalies. Detailed investigation of observations from individual sites showed that decreases in the wind speed to near-calm, and the reversal of land–sea breezes in the presence of large coastal temperature gradients, are two possible situations in which especially large NSAT anomalies may occur. Given the mid-morning timing of the 20 March 2015 eclipse, the topographical influences on NSAT anomalies in clear-sky areas may have been dominated by variations in the residual nocturnal inversion strength, as suggested by statistically significant correlations between post-sunrise temperatures and NSAT anomaly for clear-sky sites, with larger negative anomalies where post-sunrise temperatures were lower.
The MMS data used in this study are archived internally by the Met Office, and may be made available to academic researchers. Enquirers should contact the author. Gridded elevation data are available from the NOAA National Centers for Environmental Information ETOPO1 pages (http://www.ngdc.noaa.gov/mgg/global/global.html). The land use CCI data were obtained from the European Space Agency (ESA) and may be downloaded free of charge at http://www.esa-landcover-cci.org/. The eclipse variables (times of first contact, maximum obscuration, last contact and obscuration) were obtained using script available on the website of Xavier M. Jubier (http://xjubier.free.fr). Xavier is a member of the International Astronomical Union Working Group on Solar Eclipses. Data analysis in this study was performed using Python scripts and Microsoft Excel worksheets.
I declare I have no competing interests.
No funding has been received for this article.
The author would like to thank Xavier M. Jubier for provision of software used to calculate eclipse variables, and Lizzie Good for running the software in order to generate eclipse variables for each observation site.
One contribution of 16 to a theme issue ‘Atmospheric effects of solar eclipses stimulated by the 2015 UK eclipse’.
- Accepted December 10, 2015.
- © 2016 The Author(s)
Published by the Royal Society. All rights reserved.