Satellite observations of surface temperature during the March 2015 total solar eclipse

Abstract

The behaviour of remotely sensed land surface temperatures (LSTs) from the spinning-enhanced visible and infrared imager (SEVIRI) during the total solar eclipse of 20 March 2015 is analysed over Europe. LST is found to drop by up to several degrees Celcius during the eclipse, with the minimum LST occurring just after the eclipse mid-point (median=+1.5 min). The drop in LST is typically larger than the drop in near-surface air temperatures reported elsewhere, and correlates with solar obscuration (r=−0.47; larger obscuration = larger LST drop), eclipse duration (r=−0.62; longer duration = larger LST drop) and time (r=+0.37; earlier eclipse = larger LST drop). Locally, the LST drop is also correlated with vegetation (up to r=+0.6), with smaller LST drops occurring over more vegetated surfaces. The LSTs at locations near the coast and at higher elevation are also less affected by the eclipse. This study covers the largest area and uses the most observations of eclipse-induced surface temperature drops to date, and is the first full characterization of satellite LST during an eclipse (known to the author). The methods described could be applied to Geostationary Operational Environmental Satellite (GOES) LST data over North America during the August 2017 total solar eclipse.

This article is part of the themed issue ‘Atmospheric effects of solar eclipses stimulated by the 2015 UK eclipse’.

1. Introduction

A total solar eclipse occurred on 20 March 2015 with a path of totality that crossed the North Atlantic, the Faroe Islands and Svalbard, with Europe, Iceland and parts of North Africa and northern Asia experiencing a partial solar eclipse. For most of Europe, the eclipse was a morning event, with mid-eclipse occurring at a local solar time of between about 0800 in the west and 1430 in the east. A solar eclipse provides a unique opportunity to observe the Earth-system responses to an abrupt change in solar insolation. Reductions in surface temperature are most widely reported, but effects on other geophysical variables, such as surface winds, pressure and even ozone, have also been documented [1–4].

The focus of this study is on the spatial characteristics of satellite-observed surface temperature changes over Europe during the March 2015 eclipse. Other meteorological aspects of this eclipse are described in the accompanying articles in this theme issue. In particular, the reader is referred to Clark [5] for a detailed description of the changes in near-surface air temperatures (NSATs) observed in the UK. The analysis carried out here is a novel one because nearly all studies that report surface temperature changes during a particular solar eclipse event are concerned with NSAT observed at just a few stations over a limited geographical region [2,3,6–11]. Analysis of temperature changes during a single event using a large number, or high density, of observations has so far been limited to just a few studies. For example, Hanna [12] examined reports during the total solar eclipse of 11 August 1999 from 81 stations across the UK, summarizing the results on a regional basis. Clark [5] analysed NSATs from over 280 Meteorological Monitoring System stations in the UK during the March 2015 total solar eclipse. A more geographically extensive study was performed by Segal et al. [13], who analysed shelter air temperatures across the central United States (Colorado, Iowa, Illinois, Minnesota, Missouri, Oklahoma) recorded during the annular eclipse on 10 May 1994. Segal et al. [13] also looked at hourly differences in infrared temperatures recorded by the Geostationary Operational Environmental Satellite (GOES) over an area of approximately 15° longitude by 6° latitude to infer detailed spatial patterns in eclipse-induced surface temperature changes in the central USA. To the author’s knowledge, this is the only previous study to use satellite surface temperature data during a solar eclipse.

The conclusion from nearly all these studies—and, in particular, those of Hanna [12], Clark [5] and Segal et al. [13]—is that eclipse-induced temperature drops do not solely depend on the solar obscurity. Meteorology and, in particular, the cloud cover, the surface type, vegetation cover, surface moisture, topography, latitude, distance from coast, and the timing and duration of the eclipse are all thought to affect the observed temperature changes [2–4,8,12,13]. Previous studies have suggested that NSAT drops are larger over vegetated compared with bare surfaces [12,13], larger inland compared with coastal locations [3], larger at elevated sites compared with low-lying locations [3] and larger under cloud-free compared with cloudy skies [2,12]. Eclipses that occur during the earlier part of the day when temperatures are rising more rapidly are also likely to induce larger temperature drops [13]. The height of the measurement is also important, with temperature observations made closest to the ground surface exhibiting the largest reduction [1,8,11]. As a result, temperature drops reported during an eclipse vary greatly between studies.

Kameda et al. [9] provide a useful summary of temperature drops reported in other studies during a solar eclipse. They find that for air temperatures recorded at typical screen height (1.5–2.0 m) the range is 1.4 to 5.0°C with average 2.8°C for cloud-free conditions. Most of the studies reported are mid-latitude, but the 3.0°C maximum temperature drop reported by Kameda et al. [9] for their Antarctic-based study also falls within this range. However, NSAT drops that exceed this range are reported in the literature. Hanna [12]—not included in the Kameda et al. [9] summary—reports a drop of up to 6°C in the UK during the 1999 solar eclipse, while the largest NSAT drop recorded by Segal et al. [13] for the central USA during the 1994 annular eclipse is 6.4°C (at 0.15 m). Segal et al. [13] also perform model simulations of eclipse-induced temperature changes in their study and estimate a maximum NSAT drop of 7°C for mid-latitudes. By contrast, a more recent study by Eckermann et al. [14] only simulates reductions in NSAT of up to approximately 4°C over land using a numerical weather prediction model during an eclipse. The observed temperature drop also depends on the calculation method. Segal et al. [13] use two methods of calculation: (i) the temperature drop following an earlier peak (ΔT_actual) and (ii) the temperature drop compared with the estimated ‘non-eclipse’ temperature at the time of the observed minimum (ΔT_inferred). ΔT_inferred is typically 1–2°C larger than ΔT_actual, which is more commonly reported in the literature.

The objective of this paper is to investigate the factors that influence temperature drop during an eclipse. Previous studies have been unable to investigate this fully owing to scarcity of available measurements, or their limited geographical domain. Satellite data are used here to characterize the spatial patterns in temperature changes during the March 2015 solar eclipse with respect to geographical location, elevation, land use, vegetation and eclipse timing. The impact of cloud is not considered here because the satellite data used in this study are derived from infrared observations which are available for clear sky only. The benefit of using satellite data for this type of analysis is the spatial coverage and number of observations, offering an insight into surface temperature behaviour during an eclipse over a large geographical region and wide range of surface regimes. The entire European domain is analysed making this the largest observational study to date of eclipse-induced surface temperature changes (known to this author).

The temperature drop observed in satellite land surface temperature (LST) data is expected to be larger than equivalent NSAT observations as the satellite data correspond to temperature of the Earth’s ‘skin’, which often differs substantially from the corresponding NSAT. More technically, a satellite LST represents the ‘ensemble directional radiometric temperature’, which is the aggregated radiometric surface temperature of all components within the satellite field of view in the direction of observation [15–17]. LST is strongly dependent on the incoming solar radiation and responds quickly to changes in insolation. For example, near-instant (seconds to minutes) and large (several degree Celcius) changes in LST can be observed when a scene undergoes a change from bright sunshine to cloud-shadowed [18]. As a result, the LST diurnal cycle closely follows that of solar insolation, with maximum LST typically occurring just after solar noon and some hours before maximum NSAT [19–21]. Because of the sensitivity to solar heating, daytime clear-sky LSTs are typically warmer—often by several degrees Celcius—than NSAT, while night-time LSTs and NSAT are similar [18,20,22].

Although LST and NSAT may differ in magnitude, they are usually highly correlated [22–25]. Therefore, the behaviour of LST changes during a solar eclipse should also be relevant to NSAT. Segal et al. [13]—the only existing study to analyse satellite surface temperatures during the eclipse (known to this author)—observed temperature drops of up to approximately 11°C in the satellite skin temperatures during the 1994 annular eclipse of the USA, which is much larger than the NSAT changes observed in the same study (up to 6.4°C). This larger magnitude is expected, and is supported by other published studies that report changes in grass and slab temperatures, which are physically closer to LST than NSAT. For example, Hanna [12] reports reduction in grass temperatures of up to 17°C during the 1999 total solar eclipse in the UK, compared with a maximum NSAT reduction of 6°C. Kameda et al. [9] estimate a reduction in snow surface of 4.6°C compared with 3°C in NSAT in Antarctica during the 2003 total eclipse. By contrast, sub-surface eclipse temperature drops tend to be much smaller than for NSAT and become undetectable below about 20 cm depth [7,11].

Aside from the magnitude, the other significant difference between eclipse-induced LST changes compared with NSAT may be in the timing of the observed minima. NSAT minima typically occur after the eclipse mid-point, where the lag time depends on the thermal inertia of the surface [1,2,4,9]. This lag time is typically 5–30 min after mid-eclipse [1,9], and may be linearly related to the global solar radiation at fourth contact, i.e. the end of the eclipse [9]. Given the response of LST to solar insolation, eclipse LST minima are likely to occur before this since the rate of change in global solar radiation during an eclipse is nearly linear and is proportional to solar obscuration [3,9,11]. This expectation is supported by Foken et al. [7] who find that the IR surface temperature minimum measured in situ at a German site during the 1999 total solar eclipse occurred just after totality, which is 18 min before the corresponding minimum NSAT.

2. Data and methods

(a). Remotely sensed land surface temperatures

The remotely sensed LSTs used in this analysis are from the spinning-enhanced visible and infrared imager (SEVIRI) sensor, which is onboard the Meteosat Second Generation (MSG) platform. The first SEVIRI was launched on MSG-1 in August 2002, providing the operational 0-degree ‘prime’ service from January 2004. MSG-2 and MSG-3 were launched in 2005 and 2012, enabling continuity of SEVIRI observations to the present day. SEVIRI provides full-disc images every 15 min at 3 km spatial resolution at the sub-satellite point (1 km resolution for the high-resolution visible channel) [26].

Operational, near-real-time estimates of LST are produced by the Land Surface Analysis Satellite Applications Facility (LSA SAF) at the full spatial and temporal resolution of the sensor. Data from 2009 to the present day are currently available from the LSA SAF website (http://landsaf.meteo.pt/) and a real-time data feed is also available through the EUMETCAST dissemination service provided by EUMETSAT. Sensors such as the SEVIRI provide ‘top of atmosphere’ observations and geophysical parameters such as LST must be estimated, or retrieved, from these observations. SEVIRI has twelve spectral channels located in the visible and infrared regions, which includes four channels that can potentially be used for estimating LST at 3.9, 8.7, 10.8 and 12 μm [26]. The LSA SAF LSTs are derived using the ‘split window’ channels located at 10.8 and 12 μm using the generalized split window scheme proposed by Wan & Dozier [27]; further details are provided in Trigo et al. [28]. This enables the effects of the atmosphere and surface emissivity to be removed from the observations, leaving an estimate of the LST. SEVIRI LST retrievals have reported uncertainties of around 1–2 K—a result of uncertainties in the emissivity and atmospheric corrections—although larger uncertainties can occur for observations made at very large view angles or in high water vapour loading [29–32]. Biases in LST retrievals at a given location are likely to be systematic over periods of a few hours, which means that the uncertainties in the change in LST over this timescale will be small and probably much less than 0.5°C (the Noise-Equivalent Delta Temperature of the 10.8 and 12 μm channels is 0.25 and 0.37 K at 300 K, respectively; see http://www.wmo-sat.info/oscar/instruments/view/503). This is relevant for this study, where it is the change in LST that is of interest rather than the absolute temperature observed.

The largest source of uncertainty in SEVIRI LST observations is undetected cloud. LST cannot be observed through cloud and pixels that have been identified as cloudy are flagged in the LSA SAF LST datasets. However, correctly identifying cloudy satellite observations over land is non-trivial and the cloud flagging procedures sometimes miss cloud, or incorrectly identify cloud when conditions are cloud-free. Undetected cloud can cause errors in LST retrievals of several degrees Celcius and these are usually cold-biased compared with the true LST. SEVIRI LSTs flagged cloudy in the LSA SAF product are excluded from this study. For the quantitative analysis of LST time series, any pixels with more than one observation flagged cloudy in the time series are also excluded. This not only ensures that changes in LST during the eclipse are well represented, but also minimizes the risk of using cloud-contaminated observations.

The SEVIRI LST data used in this study correspond to all 15 min MSG observation slots occurring between, and including, 0715 and 1200 GMT on 20 March 2015. As a single SEVIRI scan takes around 12 min (south to north) [26], actual LST observation times are calculated for each SEVIRI pixel by taking this scan time into account.

(b). Auxiliary datasets: land cover, vegetation fraction, distance from coast and elevation

Land cover classification and vegetation information have been used in this study to investigate the spatial variation in eclipse-induced LST changes. The land cover map used here is the European Space Agency (ESA) Climate Change Initiative (CCI) 300 m land cover map for the 2008–2012 epoch (http://www.esa-landcover-cci.org/) [33,34]. These data have been used to calculate the dominant land cover class and water fraction for each SEVIRI pixel. The water fraction is used to exclude satellite LST observations for pixels with greater than or equal to 10% water, as water has a different temperature response to the land surface and may affect the results of the study.

The fraction of vegetation cover (FVC) for each SEVIRI pixel is sourced from the LSA SAF. Like LST, FVC data are available at the full spatial resolution of the SEVIRI from the LSA SAF website and EUMETCAST service. The daily data are expressed as values between 0 and 100%, with estimated uncertainties and quality information given per pixel; the overall absolute accuracy of the FVC product is expected to be within 10% for 70–75% of land pixels [35].

The earth location of each SEVIRI pixel is more-or-less constant with time. Pixel latitude, longitude and elevation for this study are sourced from the LSA SAF static dataset archive at http://landsaf.meteo.pt/. The LSA SAF land mask, available in each SEVIRI LST file, is used to calculate the distance from coast for each SEVIRI pixel.

3. Results

Figure 1 shows the eclipse parameters calculated for each SEVIRI land pixel within the European domain. Obscuration was greatest in northern Europe, with the UK, Iceland and Scandinavia experiencing over 80% obscuration. In southern Europe and the northwestern tip of Africa, solar obscuration was around 50% (figure 1a). First contact time occurred between approximately 0800 GMT in the west and approximately 1000 GMT in the east, with fourth contact occurring between approximately 0945 and approximately 1145 GMT (figure 1b,d). The timing of mid-eclipse, therefore, ranges between about 0800 (west) and 1430 (east) local solar time. Owing to the curvature of the Earth and variation in local solar time of the eclipse, the duration varies with location [13]. For this event, central and eastern Europe experienced the longest eclipse duration of over 140 min.

Figure 1.

Eclipse characteristics for land pixels within the SEVIRI European domain showing (a) obscuration, (b) eclipse duration, (c) first contact and (d) fourth contact. (Eclipse parameters obtained from Xavier Jubier, member of the International Astronomical Union Working Group on Solar Eclipses via the website http://xjubier.free.fr.)

(a). Land surface temperature evolution during the eclipse

Figure 2 shows a time series of SEVIRI LST maps over the UK from 0756 to 1041 GMT. The satellite data show that most of the UK was obscured by cloud for the duration of the eclipse. The South West and Wales had the most persistent and widespread clear skies; parts of the Midlands and east Scotland are also visible in the satellite observations although the availability of LST observations is less consistent in these regions. The South East shows some apparent cloud-free LST observations between 0911 and 0941. However, this is cloud that has been ‘missed’ by the SEVIRI cloud-screening. This is inferred from the uniformity of the LSTs in this region, where we would expect true LST to be more variable, and the apparent lack of cloud-free observations for the other time slots. In situ observations confirm that the South East was obscured by cloud during the eclipse [5]. This cloud may have been undetected as a result of the lower light levels around the peak of the eclipse, which could have affected the cloud detection scheme that includes some tests based on channels that operate at visible wavelengths.

Figure 2.

Time series of SEVIRI LST over the UK for 20 March 2015. The mean observation time (GMT) is shown on each thumbnail. Grey areas indicate missing data (non-land or cloud).

The satellite LST data show a decrease in LST from the 0911 slot, but have clearly begun to recover by the 0956 slot. Minimum LSTs are inferred to occur between the 0926 and 0941 slots, close to the timing of mid-eclipse over the UK at approximately 0930. This is supported by the single-pixel time series (Europe/North Africa) shown in figure 3, whereby LST shows an increase consistent with increasing solar insolation (and solar elevation) to the point of first contact, shortly after which, temperatures begin to decrease. Minimum LSTs occur close to mid-eclipse—recalling that the SEVIRI sampling interval is 15 min—before recovery between mid-eclipse and fourth contact. Considering all data in the analysis domain, the median time of the minimum observed LST during the eclipse occurs 1.5 min after mid-eclipse (inter-quartile range 11.0 min).

Figure 3.

LST time series for a single SEVIRI pixel in (a) northern Sweden, (b) South West UK, (c) Poland and (d) Algeria. The dark-shaded region shows the time from first contact to the eclipse mid-point, while the light-shaded region shows the time from the eclipse mid-point to fourth contact. The dotted line indicates the estimated LST change without the effects of the eclipse. The dashed line shows the inferred LST change (ΔT_inferred).

(b). Definition of land surface temperature drop

The time series in figure 3 are annotated to illustrate the two eclipse-induced temperature drop metrics used in this study: ΔT_actual and ΔT_inferred. ΔT_actual is defined as the difference between the minimum LST during the eclipse and an earlier peak. ΔT_inferred is the largest difference between an estimated non-eclipse LST and the observed LST during the eclipse. This differs slightly from the definition of Segal et al. [13] who consider the difference between the estimated non-eclipse temperature and the minimum temperature close to mid-eclipse. The relevance of ΔT_inferred is apparent in figure 3d where the LST reduces only slightly and ΔT_actual is close to zero, despite there being a very clear impact on LST. Both measures are negative where a reduction in LST has occurred.

(c). Spatial characteristics of land surface temperature drop

Figure 4 shows maps of ΔT_actual and ΔT_inferred for all cloud-free SEVIRI pixels in the analysis domain (tolerance of 1 cloudy slot over analysis period; §2). As expected, ΔT_inferred is more negative than ΔT_actual, with a median difference between the two temperature drops of −2.0°C (inter-quartile range 0.6°C). Considering ΔT_inferred (figure 4b), arguably a better metric for examining temperature drop as discussed above, smaller values generally occur in eastern regions where the local time of the eclipse is closer to solar noon and the rate of non-eclipse LST change is lower (figure 4c,d). ΔT_inferred and the local solar time of the eclipse mid-point are found to be reasonably well correlated across the whole of Europe (r=+0.37,n=68 794; not shown), with larger LST drops occurring for earlier eclipse mid-points.

Figure 4.

Change in LST during the eclipse corresponding to (a) ΔT_actual and (b) ΔT_inferred. (c) The estimated rate of change of LST without eclipse effects (see text) and (d) the time offset between the eclipse mid-point and solar noon (negative values correspond to the eclipse occurring before solar noon).

Smaller ΔT_inferred also occurs at more southerly latitudes where obscuration is lower. ΔT_inferred and solar obscuration are found to be well correlated over the European domain (r=−0.47, n=68 794; not shown). However, although nearly the full range of obscuration is encompassed by these observations (16.3–99.6%), it should be noted that a high proportion of the data occur in the 45–75% range, consistent with central and eastern Europe. The correlation with eclipse duration is stronger (r=−0.62), with longer duration resulting in a larger ΔT_inferred. Scatter in these relationships can be attributed to other effects, which are discussed in the following sections.

(d). Relationship between land surface temperature drop and vegetation/land cover

The diurnal cycle of LST is heavily dependent on vegetation cover, where the amplitude increases with decreasing vegetation cover (figure 5), so it seems logical that vegetation should also influence an eclipse-induced LST drop. For Europe as a whole, a statistically significant (p<0.05) but very weak relationship is found between ΔT_inferred and FVC (r=−0.06,n=63 027). This weak relationship is not surprising given the range of eclipse timing, solar obscurity and other factors influencing the temperature drop. Stronger relationships can be found over smaller spatial domains (table 1) where these other factors are more constant. For example, correlation coefficients of +0.31 are obtained for central Europe and eastern Europe, while stronger correlations of +0.60 and +0.52 are obtained for the Russian and central Finland domains, respectively. A weaker correlation is obtained for Italy (r=+0.14), while no statistically significant result is obtained for southwestern UK (p>0.05). However, using only the data from Wales yields a significant correlation of +0.23 (n=423), which probably reflects the sensitivity of these relationships to the nature of the spatial domain. These results indicate a positive relationship between FVC % and ΔT_inferred, where smaller LST drops are observed over more vegetated surfaces.

Figure 5.

LST as a function of time for all SEVIRI pixels within 20 km of Munich, Germany. The lines have been coloured according to the vegetation fraction for each satellite pixel. The solid red line corresponds to the pixel nominally containing Munich.

Table 1.

Sub-regions analysed in the study. The central Europe domain includes parts of Germany, Austria, Czech Republic and Switzerland, while eastern Europe includes parts of Croatia, Hungary and Slovakia. The right-hand column gives the correlation coefficient (r) between ΔT_inferred and FVC for each region and the number (n) of pixels analysed (see text); correlation coefficients where p>0.05 are shown in parentheses. The number of pixels used in the correlation may be less than the total number of cloud-free SEVIRI pixels in column 5 owing to missing FVC observations.

					correlation (FVC, ΔTinferred)
region name	latitude range (°N)	longitude range (°N)	obscuration range (%)	N cloud-free SEVIRI LST pixels	r	n
Europe	34.0, 70.0	−22.0, 40.0	16, 100	68 794	−0.06	63 027
southwestern UK	49.5, 53.0	−6.0, −2.0	86, 89	622	(−0.04)	616
central Finland	60.0, 66.0	22.0, 28.0	76, 89	1262	+0.52	225
Italy	40.0, 45.0	11.0, 15.0	47, 62	5034	+0.14	4859
central Europe	47.0, 51.0	9.0, 14.0	63, 76	8899	+0.31	8277
eastern Europe	45.0, 50.0	16.0, 20.0	52, 67	11 956	+0.31	11 806
Russia	52.0, 56.0	36.0, 40.0	47, 58	4109	+0.60	3663

Using the same sub-regions, the influence of land cover type on ΔT_inferred is assessed. Table 2 shows the median ΔT_inferred for each land cover type in each sub-region. The variation between land cover types is notable for all sub-regions apart from the southwestern UK, which has a median ΔT_inferred of −6.3°C for both land cover classes (rainfed cropland; grassland). For the other sub-regions, differences in the median ΔT_inferred between land cover classes range between 0.9°C (Finland) and 2.1°C (central Europe). For all sub-regions, the largest LST drops occur in the urban or cropland classes. There is less of a consistent pattern in land cover classes with the smallest LST drops, but for central Europe, eastern Europe and Russia, which have the largest number of land cover classes, these are generally the tree cover needle-leaved/evergreen and tree cover mixed-leaf type classes. These results are consistent with the ΔT_inferred/FVC analysis discussed above, which found smaller LST drops over more vegetated pixels. At this time of year (March), urban and cropland classes are likely to have some of the lowest vegetation cover, whereas areas covered by evergreen and mixed-leaf type forests are likely to have the highest. For central Europe, the smallest ΔT_inferred result is obtained for the bare area class—the only sub-region with this class—which appears inconsistent with this finding. However, these pixels occur at very high altitudes (greater than 2000 m) where elevation effects may dominate; this is discussed in the following section.

Table 2.

ΔT_inferred for each land use CCI land cover type for each region. Geographical domains for each sub-region are given in table 1. The mean ΔT_inferred for each region, calculated from the arithmetic average of the medians for each class, is also provided.

		median ΔTinferred (°C) [rank] (number of SEVIRI pixels)
class	description		southwestern UK	central Finland	Italy	central Europe	eastern Europe	Russia
10	cropland, rainfed	−5.2 [3]	−6.3 [1]	−4.2 [1]	−4.4 [4]	−6.6 [1]	−6.1 [1]	−3.7 [2]
		(n=45 496)	(n=36)	(n=55)	(n=3459)	(n=5308)	(n=8250)	(n=3155)
20	cropland, irrigated or post-flooding	−3.0 [11]	—	—	−4.7 [2]	—	—	—
		(n=101)			(n=27)
40	mosaic natural vegetation	−3.5 [8]	—	—	—	—	—	—
		(n=26)
60	tree cover, broadleaved, deciduous	−4.3 [6]	—	—	−4.4 [4]	−6.1 [3]	−4.8 [3]	−3.0 [3]
		(n=7615)			(n=1424)	(n=235)	(n=2188)	(n=485)
70	tree cover, needle-leaved, evergreen	−4.4 [5]	—	−3.3 [2]	—	−5.3 [6]	−4.7 [4]	−2.5 [5]
		(n=8691)		(n=1188)		(n=1944)	(n=4.7)	(n=228)
90	tree cover, mixed-leaf type	−4.3 [6]	—	—	—	−5.4 [5]	−4.6 [5]	−2.8 [4]
		(n=2382)				(n=397)	(n=503)	(n=208)
100	mosaic tree and shrub	−3.4 [9]	—	—	—	—	—	—
		(n=188)
110	mosaic herbaceous cover	−3.2 [10]	—	—	—	—	—	—
		(n=40)
120	shrubland	−2.6 [12]	—	—	—	—	—	—
		(n=305)
130	grassland	−5.7 [2]	−6.3 [1]	—	−4.6 [3]	−6.0 [4]	—	—
		(n=1602)	(n=551)		(n=36)	(n=751)
150	sparse vegetation	−3.4 [9]	—	—	—	—	—	—
		(n=1297)
180	shrub or herbaceous cover, flooded	−4.5 [4]	—	—	—	—	—	—
		(n=49)
190	urban areas	−6.0 [1]	—	—	−6.0 [1]	−6.4 [2]	−6.0[2]	−3.9 [1]
		(n=754)			(n=48)	(n=202)	(n=97)	(n=32)
200	bare areas	−3.6 [7]	—	—	—	−4.5 [7]	—	—
		(n=179)				(n=36)
mean (equal weighting)		−4.1	−6.3	−3.8	−4.8	−5.8	−5.2	−3.2

(e). Relationship between land surface temperature drop and distance from coast and elevation

For the Russia, southwestern UK and central Finland sub-domains, the range of pixel elevations is insufficient to determine any relationship between LST drop and elevation. For central Europe, eastern Europe and Italy, the variation in topography is much larger, particularly for central Europe where some cloud-free pixels occur at elevations of over 2000 m. Smaller LST drops are found at higher elevations compared with lower elevations, although the dependency appears nonlinear (figure 6a−c). For the central Europe sub-domain, the variation in temperature drop levels off above approximately 1000 m, whereas for the Italy sub-domain there seems to be little variation in LST drop with elevation below approximately 1000 m. This variation is likely to be a result of other influencing factors, such as surface type (see above), ventilation of the surface by superposition of mountain winds on the slack large-scale circulation at the time of the eclipse (http://www.willandweather.org.uk/maps/Brack2015032009.gif), and the distance from coast, which is discussed below.

Figure 6.

Variation in ΔT_inferred with elevation for (a) central Europe, (b) eastern Europe and (c) Italy, and distance from coast for pixels with elevation less than 200 m for (d) Europe, (e) central Finland and (f) Italy. The solid black line shows the 50th centile, while the dark grey region shows 25th to 75th centile limits. The light grey region shows the 5th to 95th centile limits. A minimum threshold of 20 pixels for each x-interval is required to produce these results.

For the European domain, a clear relationship between the LST drop and distance from coast is apparent, where the LST drop increases away from the coast (figure 6d); pixels with elevation >200 m have been excluded here to isolate coastal effects from any influence of elevation. This behaviour is mirrored in the southwestern UK, eastern Europe and central Europe domains (not shown), and to a lesser extent, Italy (figure 6f). However, for central Finland a positive relationship is obtained with larger LST drops at coastal locations compared with inland areas (figure 6e). The reason for this is unknown, but may be due to an inland-increase in green vegetation fraction or other local factors not explored here, as land cover type and elevation are reasonably constant over this domain. However, the results for central Finland do show an initial increase in the LST drop for very near coastal locations (5–10 km) compared with coastal locations (less than 5 km), which suggests there could still be a very localized effect in this sub-region.

4. Discussion

Previous studies have struggled to find an empirical link between temperature drop, and obscuration and other eclipse parameters, owing to the dominating influence of local factors, such as vegetation cover, land use and cloud cover, and the more limited geographical range of study and/or number of observations. In this study, the use of satellite data over a large domain has permitted this to be investigated empirically for the first time. The magnitude of LST drops during the March 2015 total solar eclipse is found to vary by several degrees Celcius across the European domain. These drops appear well correlated with the degree of solar obscuration (r=−0.47), and even more highly correlated with eclipse duration (r=−0.62), which is longer at lower latitudes and for eclipses that occur closer to solar noon (figure 1b; also see [13]). The LST drop is also reasonably well correlated with the local solar time of the eclipse mid-point (r=+0.37), which is later at more easterly longitudes (figure 4d). These findings support the earlier study by Segal et al. [13] who suggest larger temperature drops occur for earlier, and longer eclipses. For earlier eclipses, the sub-skin temperature is still cold after the night, so the skin is able to cool more rapidly when insolation is reduced. However, the correlations obtained here may not reflect the true dependency of the LST drop on these parameters, as the local circumstances of the eclipse covary. Both obscuration and eclipse duration vary with latitude and longitude (figure 1) so the correlation between LST drop and one of these factors will include effects from the other. Despite this, it is logical to expect that temperature drop should have some dependency on eclipse obscuration and timing, and model simulations [13] and the results of this study appear to support this.

As expected, the magnitude and timing of the eclipse-induced LST minima are found to differ from other studies reporting NSAT changes. The median inferred LST drop (ΔT_inferred) for the whole of Europe is −4.9°C, while a result of −3.1°C is obtained for the actual LST drop (ΔT_actual). For the southwestern UK, which is the sub-region analysed here with the highest obscuration, figures of −6.4 and −2.9°C are obtained. These LST drops are slightly larger than clear-sky NSAT drops reported in other studies (e.g. Kameda et al. [9], who find an average ΔT_actual of 2.8°C for a number of previous studies). For the UK, the values of ΔT_inferred reported here for LST are considerably larger than the values of ΔT_inferred reported by Clark [5] for NSAT, which range up to −4.2°C (median clear sky −2.2°C) during the same event. Results reported in this study also suggest the LST minima occur close to the eclipse mid-point with a median for Europe of 1.5 min after mid-eclipse. This is earlier than that expected for NSAT minima, which typically occur 5–30 min after mid-eclipse [1,9]. In practice, the timing of LST minima varies by location owing to the 15 min sampling time of SEVIRI.

The large number of observations and geographical range used in this study have enabled an investigation into geographical factors that influence LST drop, namely vegetation/land type, elevation and distance from coast. These factors are considered independently in this study, but it should be noted that they may covary, for example, vegetation and land use may covary with distance from coast. Nevertheless, the results presented here suggest each of these factors can affect the LST drop during a solar eclipse. In particular, the results presented indicate a clear relationship between vegetation fraction and LST drop, where smaller drops are obtained over more vegetated areas. Previous studies have found that the relationship between NSAT drop and vegetation has the opposite sense, i.e. larger drops are found over more vegetated areas compared with bare surfaces [12,13]. Although the drop in NSAT is related to the drop in LST, the relationship depends on the stability of the atmosphere and the properties of the surface, and in particular, the thermal roughness length. If a surface has a small roughness length, like bare soil, the temperature gradient near the ground will be sharper than over a vegetated surface, so that it is conceivable that the drop in NSAT may be significantly smaller than the drop in LST (J. M. Edwards 2016, personal communication). The behaviour of LST over different surface types during an eclipse is also reflected in the diurnal variation of LST (figure 5), where the diurnal range of LST is notably smaller over vegetated surfaces. The eclipse-induced drop in LST (and indeed the amplitude of the diurnal cycle) will depend on the health of the vegetation, which plays an essential role in controlling surface heat fluxes. The early spring timing of this eclipse corresponds to a period of low vegetation stress for most of the analysis domain, where vegetation transpiration, and therefore surface cooling, will be high. For eclipses that occur at different times of the year, in particular, when water stress is high, the relationship between vegetation fraction and LST drop could be weaker.

The results of this study suggest that the relationship between LST drop and elevation also has the opposite sense to that between NSAT drop and elevation. In a study by Founda et al. [3], who examined both model simulations and observations during an eclipse, NSAT drops were found to increase with elevation. In this study, the LST drop is found to decrease with elevation. The reason for this difference may be due to ventilation of the surface by mountain winds and greater mixing in regions with higher topographic variance, which may reduce the LST drop and conceivably increased the NSAT drop. Founda et al. [3] also found larger NSAT drops inland compared with coastal locations. In this case, LST appears to exhibit the same behaviour as NSAT, in general, although the results obtained for the central Finland sub-region, where a decrease in LST drop was found with distance from coast, suggest that other local conditions, such as vegetation changes, can dominate over coastal effects.

While the results of this study are interesting from an eclipse perspective, they are also relevant for use of LST in other applications. Applications for LST in climate and weather science applications are emerging [18,22,24,36,37] and the response of LST to changes in solar insolation over different surface types is relevant, particularly where this behaviour may have the opposite sense to that expected for NSAT.

5. Conclusion

This study describes the analysis of 15 min geostationary satellite surface temperature data obtained during the solar eclipse of 20 March 2015. The entire European domain is analysed, which corresponds to a variation in solar obscuration of between 16.2 and 99.6%, and variation in local solar time of the eclipse mid-point between around 0800 and 1430. LST drops are calculated for each cloud-free pixel based on the time series of data and are analysed with respect to solar obscuration, eclipse timing, vegetation, land cover, elevation and distance from coast. The LST drop is found to be negatively correlated with solar obscuration (r=−0.47; larger obscuration = larger LST drop) and eclipse duration (r=−0.62; longer duration = larger LST drop). Smaller LST drops are obtained for locations with a later eclipse mid-point (r=+0.37). These findings are in agreement with the earlier study of Segal et al. [13]. This study suggests the eclipse-induced drop for LST is found to be larger and occurs earlier than for NSAT, with minimum LSTs occurring close to the eclipse mid-point (median 1.5 min after mid-eclipse in this study).

At local scales, over a few degrees latitude/longitude where eclipse parameters are more consistent, the LST drop is strongly influenced by the vegetation or land cover type, where smaller LST drops are obtained over more vegetated areas. For this Northern Hemisphere eclipse in March, this corresponds to evergreen and mixed-leaf forests, while larger LST drops are obtained over urban and cropland areas, which are less vegetated at this time of year. The relationship between LST drop and vegetation has the opposite sign to the NSAT drop versus vegetation relationship found in previous studies [13], which is attributed to the variation in surface-to-lower atmosphere energy exchange with surface regime. The eclipse-induced LST drop is also found to depend on elevation, with smaller drops found in more elevated areas. Again, this relationship has the opposite sense to that for NSAT found in a previous study [3], with the difference likely to be a result of greater ventilation of the surface by winds in regions with higher topographic variance. The LST drop in this study is found to increase with distance from coast, mirroring the behaviour of NSAT established previously [3].

This study constitutes the largest analysis of eclipse-induced temperature drops to date and is the first time the behaviour of satellite surface temperatures during a solar eclipse has been properly characterized (known to this author). The methods used in this study could be applied to LST data over North America from the Geostationary Operational Environmental Satellites (GOES) during the total solar eclipse on 21 August 2017; GOES-R will be launched in 2016 and will have a sampling rate of 5–15 min. The results of this study are also relevant for other applications of satellite LST data, where knowledge of response of LST to changes in solar insolation for different surface regimes is beneficial.

Data accessibility

The SEVIRI LST, FVC, pixel geolocation and elevation data used in this study were obtained through the EUMETCAST service, but are also freely available from the LSA SAF website at http://landsaf.meteo.pt/. The Land Use CCI data were obtained from the ESA and are freely available to download from http://www.esa-landcover-cci.org/. The eclipse parameters (obscuration and timings) for each SEVIRI pixel were obtained from Xavier Jubier’s website, http://xjubier.free.fr, by batch-calling his eclipse calculator script (e.g. http://xjubier.free.fr/php/XML_xSE_LocalCircumstances.php?Ecl=+20150320&Lat=64.123456&Lng=-12.654321). Xavier Jubier is a member of the International Astronomical Union Working Group on Solar Eclipses. The analysis in this study was performed using the Interactive Data Language; the software written for this study can be requested by contacting the author.

Competing interests

I have no competing interests.

Funding

This work was supported by the Joint UK DECC/Defra Met Office Hadley Centre Climate Programme (GA01101).