Royal Society Publishing

Diagnosing eclipse-induced wind changes

S. L. Gray, R. G. Harrison


Responses in surface winds to solar eclipses have an almost mystical status but are difficult to detect in observations because of their transient nature. High spatial resolution (approx. 1.5 km grid) meteorological models now provide a new technique for their investigation. Measurements from the southern UK meteorological network during the 11 August 1999 total solar eclipse are compared with a high-resolution model ignorant of the lunar shadow's influence. Differences between the model output and measurements at the eclipse time show transient eclipse zone temperature decreases of up to 3°C, which also depressed the day's maximum temperature compared with the model prediction. Coherent responses in temperature, and wind speed and direction measurements are detected in the inland cloud-free region (from 51° to 52° N and −2° to 0° E). A mean regional wind speed decrease of 0.7 m s−1 during the maximum eclipse hour is apparent with a mean anticlockwise wind direction change of 17°; no such changes occurred in the model output. Such regional circulation changes are consistent with Clayton's 1901 cold-cored eclipse cyclone hypothesis, which may be related to the anecdotal ‘eclipse wind’.

1. Introduction

The circumstances of a total solar eclipse provide a regional experiment in which the atmospheric response to a known change in incoming solar radiation can be studied. While the solar parameters can be accurately calculated decades or centuries in advance, there is a paradox in the detailed lower atmosphere conditions only being predictable up to tens of hours before the eclipse. This mostly precludes exploiting eclipses for a specific meteorological experiment, but a retrospective analysis of measurements within the eclipse region can still be useful. For example, using this approach, Clayton (1901) deduced wind and temperature changes during the 28 May 1900 solar eclipse. Almost a century later, the 1999 eclipse of 11 August passed over well-instrumented regions of Europe, such as the UK, and consequently many observations were made (Hanna 2000). Evidence for small changes in atmospheric circulation was again apparent (Aplin & Harrison 2003), supporting previous suggestions of eclipse-related circulation changes, anecdotally referred to as the ‘eclipse wind’. Since 2003, there have been great developments in high spatial resolution atmospheric models, providing a new methodology for a more detailed regional analysis of the eclipse-induced responses in temperature, wind speed and direction.

To identify atmospheric changes potentially due to an eclipse, background weather-related changes need to be accounted for or eliminated. Previously, comparisons have been made with selected days immediately before or after the eclipse (e.g. Gross & Hense 1999), or by interpolating measurements before First Contact and after Fourth Contact to estimate the equivalent non-eclipsed conditions at the times of Second and Third Contact (where the times of the First and Fourth Contact are the first and last times when the Sun and Moon appear to touch, and the total eclipse occurs between the times of the Second and Third Contact). Neither approach is wholly satisfactory as, in the first case, day-to-day weather variability can exceed eclipse-induced changes, and in the second, the post-eclipse measurements used in the interpolation may retain an eclipse signal, such as cooling. Here, a high spatial (1.5 km horizontal gridspacing) and temporal resolution meteorological model ignorant of the eclipse and initialized from a pre-eclipse model analysis is used to predict the weather conditions at the time of the eclipse. Meteorological measurements during the eclipse are compared with those predicted by the model for the same times and positions, to account for the background weather and daily cycle meteorological changes also occurring. Differences between the observed and predicted measurements arise from eclipse-induced changes as well as due to errors in the model prediction from initial condition, boundary condition and model errors. Here, we examine both the spatial distribution of the instantaneous differences between the observations and model output and differences in the time evolution of spatial averages of the observations and model output to infer the eclipse-induced changes. We combine a regional set of meteorological measurements for the detection of the eclipse-induced changes to allow, through spatial averaging, an improved signal-to-noise ratio over the measurements made just at a single site. Through comparison of the observations with the model, we demonstrate that the observed changes would not have occurred through weather system and diurnal evolution alone.

Aplin & Harrison (2003) reviewed meteorological changes during solar eclipses, such as the often-observed reduction in surface temperature, but also wind speed and direction effects. Wind speed changes reliably reported include a substantial gust just after Fourth Contact (Anderson & Keefer 1975; Eaton et al. 1997), possibly arising from transient turbulent mixing (Winkler et al. 2001). During the August 1999 eclipse, a variation in wind direction was suggested in the observations of Aplin & Harrison (2003) at Reading and Camborne in the UK, apparently consistent with cyclonic (anticlockwise) circulation changes predicted from the cold-cored eclipse cyclone model of Clayton (1901). Prenosil (2000) also found a slight cyclonic circulation attributable to this eclipse from comparison of model simulations (with comparatively coarse, 63.5 km, horizontal gridspacing) with and without the eclipse represented. Circulation changes during the August 1999 eclipse are explored further here, using the UK's operational meteorological instrument network and the high-resolution model comparison methodology outlined earlier.

The band of totality for the 11 August 1999 eclipse extended across the far southwest of the UK, in Cornwall and some of Devon, between 1010 and 1015 UTC (Espenak & Anderson 2004). In the south and east of the UK, the eclipse was partial, for example the maximum was 97 per cent at Reading at 1019 UTC at which time the centre line of the eclipse was about 180 km to the south of Reading in the English Channel (note that local times were UTC plus 1 h for British summer time). The times of First Contact (1003 UTC at Reading) and Fourth Contact (1039 UTC at Reading) therefore indicate that the period of cooling in the southern UK lasted over an hour, on the basis of which automatic weather station measurements with at least hourly resolution have been sought.

A description of the weather conditions is given in Hanna (2000). The weather system situation early on 11 August was a weak meridionally extended ridge over the east UK with a weak low to the west (0000 UTC analysis in figure 1a; the 1200 UTC analysis is in fig. 2 of Hanna (2000)). These features moved slowly eastwards with time. The southwest of the UK, where a total eclipse occurred between 1011 and 1013 UTC, was cloudy at the time of the eclipse owing to an occluded front close to the west of Ireland; consequently, the partial eclipse over southeast England, where there was broken cloud, led to greater observed meteorological changes (Hanna 2000; Aplin & Harrison 2003).

Figure 1.

(a) Met Office synoptic analysis at 0000 UTC 11 August 1999; crown copyright, (b) Meteosat satellite image showing the lunar shadow at 1100 UTC 11 August 1999 copyright © 2001 EUMETSAT (EUropean organisation for the exploitation of METeorological SATellites) (image has been contrast enhanced), (c) locations of observation sites over southern UK with hourly weather data on 11 August 1999 (the Shinfield Park station in Reading is marked with a black square), (d) North Atlantic European domain configuration model forecast of mean sea level pressure (shaded), 1.5 m temperature (contours, interval 1 K) and wind vectors (every 10th plotted) at 1100 UTC 11 August 1999. (Online version in colour.)

2. Methodology

(a) Observations

As an example of the detailed information available, high temporal resolution data are presented from the Meteorology Department's Atmospheric Observatory at the University of Reading (51.442° N, −0.938° E). Eclipse-day wind measurements (which are newly retrieved) from a Gill Windmaster sonic anemometer at 2.85 m height sampled every 4 Hz are shown, with the temperature measurements from a cylindrical platinum resistance thermometer, with connection resistance compensation (Harrison & Rogers 2006), in a Stevenson screen at 1.25 m height sampled at 1 Hz (previously analysed in Aplin & Harrison (2003)).

To obtain the improved signal-to-noise ratio available for small changes by using multiple observations made over a region, hourly observations are analysed from the MIDAS Land Surface Stations dataset. This contains land surface observation data from the Met Office station network and was obtained from the British Atmospheric Data Centre (BADC). MIDAS data were extracted for 11 August 1999 at stations within latitudes of 50° to 53° N and longitudes of −6° to 2° E (locations shown in figure 1c) yielding 121 stations, although many stations did not have observations available at all times. The dense network of stations provides an outstanding opportunity to analyse the spatial pattern of the meteorological changes owing to the eclipse. The following types of weather stations and message reports were extracted from hourly weather observation files1 : WMO (synoptic stations selected as suitable for possible international exchange), DCNN (stations that are part of the climate network), CLBD (commercial stations that are part of the climate network) and ICAO (stations that are part of the aviation network) stations; SYNOP (synoptic), AWSHRLY (hourly synoptic observations received from an automatic climate logger reporting hourly) and METAR (synoptic observations for aviation purposes) message reports.

Wind speed and direction (at 10 m height) and air temperature (at 1.25 m height) are analysed with data reported to 1 knot, 10°, and 0.1°C resolution, respectively. Temperature was not used from the METAR reports, as it is only reported to the nearest 1°C. Wind speeds and directions are 10 min averages, from 20 to 10 min prior to the reported observation hour. There were more wind speed than wind direction observations because wind direction is undefined when the wind speed is zero. It is UK practice to regard the actual observation time as 10 min prior to the observation hour. Quality flags were checked, and data were rejected where problems were raised on the initial climate quality control process performed by the Met Office. However, to gain the greatest spatial and temporal coverage of data, data have also been included for AWSHRLY message reports without such quality control.

(b) Numerical model

The operational weather and climate forecast model used by the Met Office, the Met Office Unified Model, has been used to simulate the eclipse day from 0000 UTC to 1800 UTC. No representation of the eclipse was included in the model simulations and the model was initialized from an analysis prior to the onset of the eclipse (with no subsequent assimilation of later observations). Hence, the simulations predict how the meteorological variables would have evolved in the absence of the eclipse.

The Unified Model is an operational finite-difference model that solves the non-hydrostatic deep-atmosphere dynamical equations with a semi-implicit, semi-Lagrangian integration scheme (Davies et al. 2005). The model uses Arakawa C grid staggering in the horizontal. The vertical coordinate system is terrain-following with a hybrid-height vertical coordinate and Charney–Phillips staggering. The model can be configured either as a global model or as a limited area model with one-way nesting. In the limited area configuration, the horizontal grid is rotated in latitude/longitude to yield an approximately isotropic grid measured by Euclidean distance. The model parametrization of physical processes includes long- and short-wave radiation (Edwards & Slingo 1996), boundary-layer mixing (Lock et al. 2000), cloud microphysics and large-scale precipitation (Wilson & Ballard 1999), and convection (Gregory & Rowntree 1990).

Three nested simulations were performed. First, the model was run in its global configuration (v. 7.1, 0.4° horizontal gridspacing, 38 vertical levels) initialized from a European Centre for Medium-Range Weather Forecasts (ECMWF) analysis at 0000 UTC on 11 August 1999 (a global Unified Model analysis was no longer available, and the model has the capability to be started from an ECMWF analysis in the absence of a Unified Model analysis); the output of this run was not analysed. Next, the model was run in its North Atlantic European domain configuration (v. 7.1, 0.11° horizontal gridspacing, 38 vertical levels) using lateral boundary and initial conditions from the global model run. Finally, the model was run in its UKV (UK variable resolution) configuration (v. 7.3, gridspacing increasing from approximately 4 to 1.5 km in the horizontal and with 70 vertical levels) using lateral boundary conditions and initial conditions from the North Atlantic European domain run. The 1.5 km grid covers most of the UK and this model configuration (unlike the others) does not use a convective parametrization scheme. The UKV model timestep was 50 s and air temperatures at 1.5 m height and winds at 10 m height were output hourly over southern England and Wales for analysis. The model derives surface layer variables at standard observation heights by assuming the standard approach of the Monin–Obukhov similarity theory (e.g. Garratt 1994) to relate the gradients of model variables to the surface energy fluxes and interpolating for temperature and winds.

3. Results

The model prediction using the North Atlantic European domain configuration (figure 1d) reproduces the synoptic-scale (i.e. continental-scale weather) features described in §1 (figure 1a). At 1100 UTC, the ridge extends along the east coast of the UK, and winds are weak over southeast England (less than about 2 m s−1) and hence also Reading. Diurnal warming results in maximum 1.5 m temperatures just inland of the southeast coast at this time. The winds over Reading were weak easterlies (less than 5 m s−1) at 0900 UTC veering to weak southerlies by 1200 UTC, and temperatures over southeast England warmed from about 17°C to 19°C (290 to 292 K) between 0900 and 1200 UTC (not shown). By 1100 UTC, the time at which the eclipse effects are most evident in the results shown here, the eclipse umbra was over Romania (figure 1b shows the eclipse shadow at this time and fig. 3 of Espenak & Anderson (1997) shows the eclipse path).

(a) Comparison of high temporal resolution and hourly observations at Reading

The high temporal resolution air temperature measurements at University of Reading's Atmospheric Observatory show a minimum temperature 15 min after the eclipse maximum that was about 2°C cooler than the average temperature before and after the transient temperature depression (figure 2a). The warming rate after the temperature depression compares well with that occurring due to sunrise on the same morning when the skies were also relatively clear of cloud over Reading (not shown). Hourly surface meteorological station data from the MIDAS station closest to Reading University are compared with the high temporal resolution measurements. This MIDAS station is Shinfield Park (station 831, AWSHRLY message report) at 51.4175° N, −0.95154° E, about 3 km from the University observatory. The comparison with the Reading raw data also illustrates the variability that is concealed in making hourly measurements. The hourly measurements are consistent with the high temporal resolution measurements and also clearly show a response to the eclipse; the temperature at 1100 UTC is 1.5°C cooler than that from a linear interpolation between 0900 and 1200 UTC (Reading is already in the penumbra of the eclipse at 1000 UTC, so this time is excluded from the interpolation).

Figure 2.

Observations at Reading from hourly data at Shinfield Park (MIDAS dataset, squares plotted at the actual observation time of 10 min prior to the observation hour) and high temporal resolution measurements at Reading University's Atmospheric Observatory (thin lines) for (a) temperature (1 Hz sampling frequency smoothed to 10 s averages and 5 min moving averages in grey and black, respectively), (b) wind speed (4 Hz sampling) and (c) wind direction (4 Hz sampling). The times of First Contact, maximum eclipse and Fourth Contact at Reading are marked by dashed black lines. Hourly observations are for 1.25 m screen temperatures and 10 m winds. High temporal resolution temperature measurements are 1.25 m screen temperatures (identical data to that in fig. 4a of Aplin & Harrison (2003)); the high temporal resolution wind measurements are from a sonic anemometer at 2.85 m height. (Online version in colour.)

The rapidly sampled wind measurements are consistent with the synoptic-scale evolution with light winds that are generally easterly before 1100 UTC (from a wind direction of around 90°) veering to southerly at about 1130 UTC. The winds are particularly light with relatively small temporal variability between about 0930 UTC and 1100 UTC leading to fluctuating wind direction data. The veering wind evolution is also evident in the 3 m cup anemometer measurements taken at Reading (fig. 7 of Aplin & Harrison (2003)) occurring during the eclipse period from about 1100 to 1200 UTC. Aplin & Harrison (2003) also reported a pronounced drop in wind speed and a reduction in variability. The hourly wind speed measurements from Shinfield Park have values towards the upper limit of the high temporal resolution measurements (likely a consequence of the difference in measurement height) but the pattern of temporal evolution is well captured, with reduced wind speeds at 1000 and 1100 UTC relative to those at earlier and later times. The hourly wind direction measurements are also consistent with those at high temporal resolution. A veer in the wind direction occurs at 1000 UTC followed by a backing at 1100 UTC before veering again at 1200 UTC.

(b) Comparison of the observations and model output

A spatial comparison between the MIDAS observations and UKV model output is shown in figure 3 at 0900, 1100 and 1300 UTC. The temperature difference between the observations and model output (left column) is evenly distributed between positive and negative values with magnitudes generally less than 2°C at 0900 UTC, is negative over most of the domain down to −3°C at 1100 UTC and is still generally negative over the east of the domain down to −3°C at 1300 UTC. Hence, the effect of the eclipse in the observations at 1100 UTC is clear (far exceeding typical errors in short-range model prediction) and temperatures remain depressed relative to the model at 1300 UTC (hence the diurnal temperature maximum is depressed by the eclipse). Wind vectors from both the observations and model output (right column) show the anticyclonic circulation over eastern England associated with the synoptic-scale ridge. However, there are local differences. These are particularly evident at the land–sea transitions where local coastal effects such as sea breezes (Simpson 1994), which may not be well represented in the model, could be important. A clear effect of the eclipse is not obvious in wind vectors in these plots; the primary response is in temperature.

Figure 3.

MIDAS observations of temperature minus UKV model temperature interpolated to station locations, bi-linearly interpolated onto a regularly spaced 40×40 point grid for contouring (°C with 0°C isotherm marked, left column) and MIDAS wind vectors (black) and UKV model wind vectors interpolated to station locations (grey) (right column) at (a,b) 0900 UTC, (c,d) 1100 UTC and (e,f) 1300 UTC. Note that wind vectors are shown only for the UKV model output at times when MIDAS observations are available. The inland region used for spatial averaging is denoted by a blue square within which the location of the Shinfield Park station in Reading is marked. (Online version in colour.)

The eclipse signal in the temperature and wind fields can be extracted by performing local spatial averages. Data at stations within a box from 51° to 52° N and −2° to 0° E are shown from the observations (left column) and model output (right column) in figure 4. This region was chosen because it is inland, has a large density of observations, is the region of strongest cooling due to the eclipse because it is relatively cloud-free (figure 3c) and it includes Reading. The model output is a prediction of how these fields would have changed through weather system and diurnal evolution alone.

Figure 4.

Temporal evolution of MIDAS surface meteorological observations within a box from 51° to 52° N and −2° to 0° E (left column, plotted at the observation hour) and UKV model output interpolated to the station locations at times when MIDAS observations are available (right column) for (a,b) temperature, (c,d) wind speed and (d,e) wind direction. Thin lines are for individual stations. Thick black line is the mean value with error bars given by ±1.96×standard error (yielding the 95 per cent confidence interval). (Online version in colour.)

The model predictions of temperature and winds are generally consistent with the observations prior to the eclipse onset (at and before 0900 UTC) although the warming expected to be associated with the usual diurnal variation appears slightly slower in the model than that observed averaged over this particular region. Hence, the model prediction provides a reliable representation of the expected changes in temperature and winds due to the background weather system and diurnal evolution averaged over this region. The contemporaneous and diurnal peak cooling due to the eclipse are clearly evident from comparison of figure 4a,b; the synchronous cooling is largest at 1100 UTC (a cooling of about 1°C relative to a temperature interpolated from those at 0900 and 1200 UTC) but is possibly also present at 1000 UTC. For the observed wind speeds, the spread among the stations and temporal variability appears large; this is partly due to the relatively weak winds and the reporting resolution of 1 knot (approx. 0.5 m s−1; figure 4c). However, a relatively large spread among the stations is also found in the model output (figure 4d), implying that local effects such as station altitude and boundary-layer properties also contribute to the spread. Despite this, weaker observed region-mean wind speeds occur at 1000 and 1100 UTC and the weakening at 1100 UTC, relative to a wind speed interpolated from those at 0900 and 1200 UTC, of about 0.7 m s−1 is significant. Significant is defined here as significantly different (outside the 95 per cent confidence interval of the changed mean, implying that it exceeds the natural variability) to a value at the same time obtained by interpolating from values before and after the eclipse; the justification for the comparison of the eclipse-mean against a linearly interpolated value of the observations is from the model output, which clearly shows a steady progression with time for the same day without an eclipse. As no appreciable wind speed reduction is found in the model output at this time (figure 4d), the observed weakening can be attributed to the eclipse. The gradual veering of the winds due to the eastwards progression of the synoptic-scale ridge can be seen in the mean wind directions from both the observations and model output (figure 4e,f). However, a transient significant backing of the mean winds (from an interpolated value of 154° to the actual value of 137°) at 1100 UTC is only evident in the observations implying that this can also be attributed to the eclipse.

4. Discussion and conclusions

Atmospheric changes attributable to the eclipse of 11 August 1999 have been derived from comparison of hourly observations from meteorological stations with predictions from a high temporal and spatial resolution meteorological model ignorant of the eclipse. Comparison of high temporal resolution and hourly observations at Reading demonstrate that the meteorological signal of the eclipse (cooling, wind speeds and wind variability, and changes in wind direction) is discernible in hourly observations as well as those at higher temporal resolution. Comparison of hourly observations from meteorological stations with the model output over southern England demonstrate a clear contemporaneous cooling and diurnal peak temperature reduction but the wind changes are less clear. Comparison of mean values over those stations in a 2° in longitude by 1° in latitude region, which includes Reading, additionally reveals a significant weakening in wind speed (of about 0.7 m s−1) and backing of the wind direction (from about 154° to 137°) that is not evident in the model output and so attributable to the eclipse.

The backing of the winds attributable to the eclipse implies eclipse-induced easterly winds. This is consistent with a cyclonic circulation centred to the south and hence with the cold-cored eclipse cyclone model of Clayton (1901), given that the eclipse umbra was over Romania at the time of the signals in the hourly observations. The much coarser-resolution model simulation comparison by Prenosil (2000) found an eclipse-induced cyclonic circulation over Northern France, Benelux and Germany at 1030 UTC, and interpreted it as the eclipse cyclone; a cyclonic circulation over the Irish sea at this time was attributed to an eclipse-induced impact on the sea-breeze circulation (also seen in the model simulations of Gross & Hense (1999)). The inland region easterly eclipse-induced winds found here are consistent with fig. 10 of Prenosil (2000). Aplin & Harrison (2003) also inferred a cyclonic circulation during the 11 August 1999 eclipse but from high temporal resolution point observations at Reading and Camborne.

Comparing observations from a dense meteorological station network in the eclipsed region with a high-resolution weather model forecast of the non-eclipse meteorological conditions provides a new basis to investigate the atmospheric response; such high-resolution models were not available at the time of the eclipse. The averaged response in the inland region of maximum eclipse supports the hypothesis of eclipse-induced atmospheric circulation changes. While the anecdotally reported ‘eclipse wind’ itself is not a sufficiently clearly defined phenomenon to allow its properties to be identified, the surface wind changes clearly detected in this work may contribute to an individual's memory of viewing a solar eclipse.


M. Provod undertook an important preliminary study of the meteorological data available. Dr K. L. Aplin provided the high-resolution Reading Observatory temperature data and helpful comments. The ultrasonic anemometer measurements were supervised by Dr J. F. Barlow. We thank the Met Office for making the Met Office Unified Model and MIDAS dataset available, the latter through the British Atmospheric Data Centre (BADC). We thank the Natural Environment Research Council (NERC) Centre for Atmospheric Science Computational Modelling Services for providing computing and technical support and diagnostic facilities for the use of the model.


  • Received January 4, 2012.
  • Accepted February 13, 2012.


View Abstract