Professional Documents
Culture Documents
Metz Assessing The Impact of A Solar Eclipse On Weather and Photovoltaic Production 85364
Metz Assessing The Impact of A Solar Eclipse On Weather and Photovoltaic Production 85364
net/publication/287208258
CITATIONS READS
8 1,549
9 authors, including:
Some of the authors of this publication are also working on these related projects:
Eweline: Development of innovative weather and power forecast models for the grid integration of weather dependent energy sources View project
All content following this page was uploaded by Carmen Köhler on 18 December 2015.
Abstract
With the strong expansion of the installed renewable energy over the last years, the relevance of weather
forecasts for operating the German power system has considerably increased. In that context, rare but
important events like the solar eclipse on the morning of 20 March 2015 pose an additional challenge when
operating the power system, as it affects the photovoltaic (PV) power production by inducing strong gradients
in the feed-in. In order to maintain grid stability, the uncertainties associated with the eclipse have been
estimated in advance for planning necessary precautions. Especially the maximum gradients in PV-power
were of importance for the provision of balancing energy. Numerical weather prediction (NWP) is very suited
for this assessment, as it allows to consider the complex mechanisms occurring in the atmosphere. Thus the
impact of the eclipse on meteorological parameters which affect the PV-power generation were evaluated.
Sensitivity studies with NWP models have been conducted in order to assess the reduction in short wave
radiation and temperature during the total solar eclipse months before the actual event. For this purpose,
model simulations with the non-hydrostatic COSMO models from the German Weather Service (DWD) have
been performed over Germany and Europe. As the weather situation and especially the cloud cover during the
eclipse could not be known in advance, a realistic worst case (clear sky conditions) and a best case (overcast
conditions) scenario were simulated over Germany. Thereof the PV-power production has been estimated and
analyzed for the different scenarios. The NWP model data from the sensitivity studies are openly distributed
(doi:10.1594/PANGAEA.839163). As near real-time NWP simulations considering the solar eclipse were
conducted a few days prior to the event, they are herein validated with measurements. Furthermore, the actual
PV-power production and actions taken by the TSOs during the solar eclipse are stated.
Keywords: solar eclipse, short wave radiation, photovoltaic power, numerical weather prediction
Germany, Ahrens et al. (2001) could document the im- in power forecasts resulting from weather phenomena
pact of the solar eclipse on the lower planetary bound- must be accounted for. For PV-power production, a so-
ary layer by means of routine observations. Vogel et al. lar eclipse represents such a challenging event (Morris,
(2001) studied the solar eclipse and its influences on 2014). The total eclipse over Europe during the morn-
temperature and wind speed in the Upper-Rhine Val- ing of 20 March 2015 was anticipated with great ap-
ley with a non-hydrostatic model in which clouds were prehension as it led to high gradients in power produc-
eliminated. Gross and Hense (1999) used the hydro- tion. To ensure grid stability for this day, a long-term
static local forecast model that was at that time oper- timetable was strictly adhered to by the TSOs. The early
ational at the German Weather Service to review the estimation of the possible maximum gradients was es-
effects of the total solar eclipse on mesoscale circula- pecially crucial for elaboration of measures. Thus stud-
tion. Instead of constraining model physics, they chose ies for investigating possible effects of the solar eclipse
a differing date that provided the desired maximum in- in 2015 on the production of solar electricity in Ger-
solation for assessing maximal influences of the solar many have been conducted in advance, e.g., Weniger
eclipse. As the forecast quality of NWP models is im- et al. (2014); Weinhold (2014); Köhler et al. (2015);
proving progressively, forecasted weather conditions are Kreifels et al. (2015). The study by Weniger et al.
also used as reference to assess the impact of a so- (2014) highlighted the chances of large gradients in ac-
lar eclipse. Gray and Harrison (2012), for instance, cumulated PV-power in Germany. Power gradients on a
used observations from the 1999 eclipse in the southern sunny day were predicted to increase by 3.5 times during
United Kingdom and compared these with the output of the eclipse (Weniger et al. (2014); Weinhold (2014)).
a high resolution NWP model ignorant of the eclipse. For the regional scale of distribution system operators
A well documented solar eclipse outside of Europe (DSO), Köhler et al. (2015) studied the losses in load
occurred on 15 January 2010 over India under favor- flows in the distribution grids of a testing ground during
able weather conditions. Ratnam et al. (2010) and Sub- the solar eclipse, using the here presented NWP model
rahamanyam et al. (2011) compare surface observa- simulation data. They point out that local gradients are
tions during the eclipse with observations from preced- in the same magnitude as experienced on cloudy days
ing and subsequent days. The effect of the solar eclipse and thus are not critical on the smaller scale of DSOs.
on the sea breeze were studied in Subrahamanyam The cloud cover posed one of the main challenges in
and Anurose (2011). Kadygrov et al. (2013) sum- predicting the effects of the solar eclipse. As the clouds
marize measurements taken at three different locations can not be predicted months or even weeks in advance,
in Russia concerning three solar eclipses, respectively assumptions had to be made which cover all aspects of
(29 March 2006, 1 August 2008 and 4 January 2011). In possible solar radiation impacts. Additionally, the effect
the United States (US) scientists and amateurs are cur- of the solar eclipse on ground temperature is of impor-
rently planning and preparing for the total solar eclipse tance for the PV-power industry, as lower temperatures
that will cross the US from north-west to south-east on cause an increase in PV-panel efficiency (Kaldellis
21 August 2017. The American Astronomical Society et al., 2014). The study of Weniger et al. (2014) pro-
is organizing special workshops for this occasion, see vided a very good insight on the potential effects of the
Arndt et al. (2012). solar eclipse on the power system. Therein, as in com-
The last intensively studied total solar eclipse over mon practice, several simplifications were adopted, i.e.
Europe was on 29 March 2006. The open access jour- the effect of the eclipse on the weather has been ne-
nal Atmospheric Chemistry and Physics dedicated a spe- glected and the reduction in irradiation caused by the
cial issue to this event (Zerefos et al., 2007). Therein, eclipse has been assumed linear with the obscuration of
Gerasopoulos et al. (2008) gives a multi-disciplinary the sun. Also, the temperature reduction and other atmo-
overview of the conducted measurements and findings, spheric feedback mechanisms as induced by the eclipse
including eclipse induced meteorological effects, but are often neglected. Given the lack of knowledge on the
also effects on crops or marine life. Founda et al. (2007) effect of an eclipse on PV-power production, a calcu-
summarizes the effects of this total solar eclipse on me- lation taking the effects of the eclipse on the weather
teorological variables across four stations in Greece. explicitly into account is meaningful in order to reduce
Additionally, the results of numerical simulations with the overall uncertainty of this assessment. For these rea-
the Weather Research and Forecast (WRF) model are sons, and also because the operational working chain of
compared with observations. For this purpose, the WRF the system operators is based on NWP model data, the
model was run twice, once accounting for the solar German Weather Service conducted NWP model sim-
eclipse and once without obscuration of the sun. Emde ulations for the total solar eclipse well in advance for
and Mayer (2007) present results of a fully three dimen- best and worst case situations (Köhler et al., 2014). The
sional simulation of the radiation transfer during this to- openly published data and insights were used by various
tal solar eclipse. Also Kazantzidis et al. (2007) studied power forecast providers as well as for an eclipse as-
the attenuation of global irradiance over Greece. sessment of the Fraunhofer Institute for Solar Energy
In recent years the contribution of renewable en- Systems ISE (Kreifels et al., 2015). A detailed data
ergy has gained in importance with regard to the power description along with a retrospective look at the solar
mix. In order to maintain grid stability, uncertainties eclipse on the 20 March 2015 is presented in this work.
Meteorol. Z., PrePub Article, 2015 C. Köhler et al.: Impact of a solar eclipse on weather and photovoltaic production 3
12
0.25
The path of the total solar eclipse and the location of wave length [nm]
Figure 4: COSMO-EU with the cloud free setup at 9:45 UTC. On the left is the short wave radiation under clear sky conditions for the
COSMO-EU, in the middle for the total solar eclipse COSMO-EU run, and on the right is the difference between the eclipse and operational
COSMO-EU run.
at 9:45 UTC, when the solar eclipse reaches its peak. 6.1.1 Worst case weather situation – 20 March
Regarding only the grid points above land, the mean dif- 2014
ference is −182.36 Wm−2 , resulting in a reduction of the
incoming solar radiation of 31 %. The effect of the two The worst case situation is a good example of anti-
opposite movements of the umbra and the sun is clearly cyclonic south-westerly flow (SWA). In Figure 5 the
visible in the COSMO-EU domain plotted in Figure 4. large-scale weather situation is visualized for this case.
The ellipse of maximum impact is directly centered over A vigorous low pressure system lies just to the south of
Germany. Iceland, helping to advect a substantial body of warm
air towards Central Europe, where a pronounced upper
6 COSMO-DE sensitivity studies ridge has formed, moving gradually east during the day.
The COSMO-DE sensitivity studies were also con- A large surface high is centered over the Balkans, ex-
ducted in summer 2014, well before the actual solar tending its influence to most of Germany, while frontal
eclipse event. The studies were aimed to assess the pos- systems associated with the Icelandic low are cross-
sible maximal and minimal impact of the solar eclipse ing the British Isles and reached western France by
in 2015 on surface weather conditions and especially the evening. Over Germany, the strongly anticyclonic
on solar radiation over Germany. A clear sky day with nature of the flow, combined with sufficient insolation
appropriate astronomical conditions was chosen as the at the spring equinox and a relatively dry troposphere
worst case scenario for a realistic assessment of the max- helps to break up any remaining layers of stratiform
imal reduction of solar radiation at the ground over Ger- cloud quickly during the daytime, leading to largely
many: 20 March 2014. For comparison, a best case sim- clear skies.
ulation is produced for a cloudy day, where the influence
of the solar eclipse is expected to be minimal: 16 March 6.1.2 Best case weather situation – 16 March 2014
2014. For each case, the COSMO-EU and subsequently
the COSMO-DE was run with and without the solar Over Central Europe a strong north-westerly flow has
eclipse, respectively. The effect of the solar eclipse was formed between a high pressure system over Biscay and
then estimated by comparing the two model runs. Based a large trough over Scandinavia. Germany is situated in
on this study, these two days were equally used as best a wide warm sector in which an originally warm and
and worst cases in Kreifels et al. (2015), for estimating moist air mass has been advected around the Atlantic
aggregated PV-power. ridge with a long track over relatively cold sea surfaces.
This warm advection results in the formation of an inver-
6.1 Weather situations sion layer in approximately two kilometers height above
The large-scale weather situations for the case studies ground, which allows for low stratus clouds to persist
are summarized in the following sections. For illustrat- over the whole day. The clouds only break up in the far
ing purposes, the objective weather pattern classification south-west and westernmost parts of Germany, aided by
method of James (2011) is used, which is based on the lee-effects from neighboring hilly regions. Figure 6 il-
Hess and Brezowsky Grosswetterlagen (e.g. Hess and lustrates the large scale weather situation for this best
Brezowsky (1977)). case simulation.
6 C. Köhler et al.: Impact of a solar eclipse on weather and photovoltaic production Meteorol. Z., PrePub Article, 2015
Figure 5: Visualization of the large scale weather situation for 20 March 2014 12:00 UTC, which can be categorized as an anticyclonic
south-westerly pattern. On the left hand side, the mean sea level pressure in hPa (contour lines) and the precipitable water content (contour-
filled field) are visualized. On the right hand side, the contour lines symbolize the geopotential of the 500 hPa level in dam and the shading
indicates the relative topography between the 500 and the 1000 hPa level (provided by P. James, DWD).
Figure 6: As Figure 5, but for the best case simulation of 16 March 2014 12:00 UTC. Subjectively, this weather pattern can be classified as
a cyclonic north-westerly pattern.
6.2 Influence of the solar eclipse time lag, the modifications in the surface energy budget
and near surface temperature effect the whole boundary
Although the effects of the solar eclipse depend strongly layer through changes in the vertical transport of energy.
on the local weather conditions and the time of year, the During this process, the temperature profile is modified,
general impact on the lower troposphere can be summa- which in turn impacts other variables such as wind and
rized as follows. In daytime and cloud free conditions, humidity. Noticeable intensification may be observed
the most pronounced effect is on the surface energy flux on (cool downslope) valley winds (Vogel et al., 2001)
due to the dramatic reduction of incoming solar radia- but also on larger scale wind circulations (Gross and
tion. Temperature close to the ground drops as a direct Hense, 1999). Depending on local conditions such as
response to the reduced surface energy flux. The lack of orography or soil water content, the formation or dis-
energy input cannot be compensated during the rest of solution of clouds may be affected (Anderson, 1999).
the day and the temperature remains lower than it would Even low-frequency gravity waves caused by the travel
under undisturbed (non-eclipse) conditions. With some of the moon’s shadow at supersonic velocity can be ob-
Meteorol. Z., PrePub Article, 2015 C. Köhler et al.: Impact of a solar eclipse on weather and photovoltaic production 7
Figure 7: Comparison of the short wave radiation from the reference COSMO-DE run (left) and the COSMO-DE run considering the solar
eclipse (middle). On the right is the difference between the model runs. The initialization time for the model simulation is 20 March 2014
00:00 UTC. The forecast for 09:45 UTC is shown.
Figure 9: Comparison of the two meter temperature from the reference COSMO-DE run (left) and the COSMO-DE run considering the
solar eclipse (in the middle). On the right is the difference between both. For the model runs, the initialization time is 20 March 2014
00:00 UTC and the 10 hour forecasts are shown.
perature also lags behind the signal in short wave radi- wave radiation above land points within the COSMO-
ation. Whereas the maximal reduction of solar radiation DE domain lies at 148 Wm−2 . Yet this yields a relative
over the whole model domain is observed at 9:45 UTC reduction of 63 %. The short wave radiation over Ger-
(see Figure 7), the maximal mean reduction in two meter many for the COSMO-DE run from 16 March 2014 is
temperature is observed in the subsequent model output depicted in Figure 10. The reference COSMO-DE run is
for 10:00 UTC. on the left, the COSMO-DE run with the solar eclipse is
Figure 9 illustrates the spatial variability in the in the middle and the difference of both plots is shown
eclipse-induced deficit of two meter temperature at on the right of Figure 10. While Germany mostly lies un-
10:00 UTC. The mean difference for all land points is der thick low stratus clouds, the cloud cover breaks up in
−2.67 °C. For a small domain centered around Athens, the south-western part, see Section 6.1.2. Thus, the so-
Founda et al. (2007) found a modeled average tem- lar eclipse reduces the incoming radiation by as much as
perature response of −2.7 °C for land areas as conse- 404 Wm−2 in the south-west, while reductions of only
quence of the eclipse in March 2006. Gray and Har- 4 Wm−2 are possible in the north-east. The same can
rison (2012) compared temperature observations dur- be observed for the temperature near the ground. Only
ing the eclipse in August 1999 in the United Kingdom in the southwestern part of Germany, where the clouds
with temperature forecasts on a 1.5 km grid from the Met break up, is a reduction in temperature of up to 2 °C ob-
Office Unified Model which did not consider the solar served. For all land points within the model domain, the
eclipse. They confirmed temperature decreases of up to mean deviation in two meter temperature is −0.91 °C at
3 °C in the transient eclipse zone. Due to cloudy condi- 10:00 UTC.
tions, their results reveal only a damped influence of the Due to the thick cloud layer in the north, Falkenberg
solar eclipse on two meter temperature. The amplitude only has a peak in solar radiation of 40.69 Wm−2 in the
of temperature reduction strongly depends on local in- reference run. The maximum difference to the eclipse
fluences, such as the local weather situation (e.g. cloudi- model simulation occurs during the afternoon due to al-
ness, advection), albedo, vegetation, or soil moisture. In terations in the cloud cover because of feedback mech-
the alpine region, the two meter temperature barely re- anisms. Nevertheless, the influence of the eclipse on the
sponds to the solar eclipse (see Figure 9). This is due to two meter temperature is negligible (not pictured).
the high albedo and insulating effects of the snow cover
which is simulated by the model.
Figure 10: Comparison of the short wave radiation from the reference COSMO-DE run (left) and the COSMO-DE run considering the solar
eclipse (middle). On the right, the difference between both model runs is shown. The initialization time for the simulations is 16 March 2014
00:00 UTC, the 9:45 UTC forecast is shown.
COSMO-DE
tovoltaic plants. In this respect, an analysis of the ef- 0.6
Worst
fect of the solar eclipse on the irradiance, temperature 0.5
case
and clouds is very important for system operators. The
arising effect on the total PV-power production has to 0.4
20
PV-power is then estimated by a probability weighted
average of power corresponding to the different param- 15
eter combinations. Finally, the results are scaled to the
installed capacity and regionally aggregated.
10
The PV-power production estimated for Hamburg
(arbitrarily chosen) is given in Figure 11a. The normal-
5
ized PV-power is displayed as a function of time for
Best
the best and worst case COSMO-DE model runs with case
0
and without the solar eclipse. To assess the total effect 00:00 06:00 12:00 18:00 00:00
of the solar eclipse on the German power system, PV- Time [UTC]
power time series have been aggregated over the whole Figure 11: PV-power production estimated for a) Hamburg and
of the country. The results are displayed in Figure 11b. b) Germany on the basis of the COSMO-DE simulations initialized
Assuming an installed capacity of 37 GWp, a maximal on the 20 March 2014 (worst case) and 16 March 2014 (best case).
reduction resulting from the solar eclipse of 18 GW and The results are shown for Germany in the operational COSMO-DE
2.9 GW is observed at 9:45 UTC for the worst and best (dashed blue line) and the COSMO-DE run with solar eclipse (solid
case, respectively. In addition, the solar eclipse causes a red line). The effect of the solar eclipse is marked by a light blue
maximal power gradient of 6.20 GW (15 min)−1 (worst area.
case) and 1.12 GW (15 min)−1 (best case) between 10:00
10 C. Köhler et al.: Impact of a solar eclipse on weather and photovoltaic production Meteorol. Z., PrePub Article, 2015
Figure 12: Percentage of difference in irradiance between the non-eclipse and eclipse COSMO-DE model run for the worst (left) and best
(middle) cases along with their difference (right) for 10:15 UTC forecasts.
and 10:15 UTC. The maximum hourly gradient occurs 9 Managing the actual solar eclipse
between 9:45 and 10:45 UTC and yields 19.9 GW h−1 .
These figures clearly show the relevance of the solar The real weather situation was better assessable the
eclipse for the power system operation. week before the actual solar eclipse occurrence. Preced-
Additionally, in the best case scenario, a small reduc- ingly, model runs with the COSMO-EU and COSMO-
tion in the aggregated PV-power for Germany is visible DE were conducted accounting for the eclipse. The
in Figure 11b well after the eclipse. This reduction can model data was provided via an ftp server to the open
be attributed to the change in cloud cover induced by public. A meteorological analysis, challenges for the
the lack of energy input. As summarized in Anderson TSOs, as well as the concerted actions amongst the four
(1999), the response of clouds on the eclipse are versa- TSOs and support by the DWD are stated in the follow-
tile: while stratocumulus often increase horizontally and ing.
vertically, small scale convective clouds tend to dissi-
pate. These effects are clearly visible in Figure 12. By
comparing the COSMO-DE eclipse model runs to their 9.1 Meteorologist’s point of view
reference runs, the reduction of the short wave radiation
in percent is computed for the worst (left) and best (mid- On the 20 March 2015, Germany was influenced by an
dle) case scenario. Their difference (right hand side in upper air ridge with its center extending from Great
Figure 12) highlights the regions where the changes in Britain to northern Germany. During the preceding
cloud cover occurred. Thus, not using a physical NWP night, low stratus clouds had formed in the western and
model for the PV-power calculation may lead to er- north-western parts of the country supported by the ad-
rors with a magnitude of 20 % in areas with significant vection of moist air from the North Sea. Patches of this
change in cloud cover (see west Germany, France). low level cloud cover were able to persist throughout the
whole day, whereas elsewhere clear sky conditions pre-
vailed. Only in the late afternoon northern Germany was
8 Data description influenced by an intensive upper level shortwave trough
The COSMO model data for the presented sensitivity approaching the North Sea. The weather situation during
studies is fully available. The data was published by the solar eclipse is very similar to the worst case sce-
PANGAEA, Data Publisher for Earth & Environmen- nario and led to comparable gradients in PV production,
tal Science, see Köhler et al. (2014). The supplemen- see Section 7.
tary data and further documentation are available at The extent of the forecasted low cloud cover fields
http://dx.doi.org/10.1594/PANGAEA.839163. The files for the day of the eclipse varied during the model runs
are published in their original GRIB format and con- over the week before the actual eclipse. Whereas a clear
tain the model initial conditions, constants and fore- signal was present at the beginning of the week, the pre-
cast variables. The variable short names are specified dicted low cloud cover was not as distinct approaching
in the model and database documentations by Schulz the 20 March 2015. Additionally to NWP model fore-
and Schättler (2011) for the COSMO-EU data and by casts, special weather reports were issued by the DWD
Baldauf et al. (2011a) for the COSMO-DE. Boundary synopticians of the operational forecast service. Having
conditions for the model simulations can be made avail- the low stratus clouds in the north-west, the numerical
able on request. weather prediction in the control area of Amprion was
Meteorol. Z., PrePub Article, 2015 C. Köhler et al.: Impact of a solar eclipse on weather and photovoltaic production 11
Figure 13: Comparison of the short wave radiation from the reference COSMO-DE run (left) and the operational COSMO-DE run
considering the solar eclipse (middle). On the right is the difference between the model runs. The initialization time for the model simulation
is 20 March 2015 00:00 UTC. The forecast for 09:45 UTC is shown.
Figure 14: Comparison of the low cloud cover from the reference COSMO-DE run (left) and the solar eclipse COSMO-DE run (middle). On
the right are the observed clouds based on the NWCSAF cloud classification. The initialization time for the model simulation is 20 March
2015 00:00 UTC. The forecast for 15:00 UTC is shown.
a challenge, while clear sky conditions dominated in the ble in Figure 14 in which the low cloud cover of the op-
control areas of 50 Hertz, TenneT and TransnetBW. erational COSMO-DE (left) and the eclipse model run
The short wave radiation at the surface is shown (middle) is shown for 15:00 UTC. As reference, a satel-
in Figure 13 for the model run of the 20 March 2015 lite based cloud classification product from Nowcasting
00:00 UTC and the forecast for 9:45 UTC. The op- Satellite Facility (NWCSAF, see Derrien et al. (2013))
erational COSMO-DE run, the COSMO-DE run ac- is depicted on the right hand side in Figure 14. The in-
counting for the solar eclipse and their difference are crease in low cloud cover is 26 % at 15:00 UTC and the
depicted on the left, middle and right hand side, re- clouds continued to intensify. This persistence and evo-
spectively. The solar eclipse causes a mean reduction in lution of the low cloud cover is consistent with the satel-
solar radiation over land of 336.06 Wm−2 (68 %). This lite information for low and very low stratus clouds.
is consistent with the worst case simulations in Sec- After the solar eclipse, the Lindenberg Meteoro-
tion 6.2.1. The maximal reduction in short wave radi- logical Observatory (Richard Aßmann Observatory) of
ation is 438.56 Wm−2 . the DWD evaluated the conducted COSMO forecasts
The low clouds which formed in the north-east differ against measurements from their Falkenberg site which
when regarding the two model runs. As stated in An- are used and designed to monitor climate and environ-
derson (1999) and discussed previously, the low cloud mental processes. In Figure 15, the model runs with the
cover is intensified due to the lack of solar radiation in operational COSMO-DE setup (dashed blue line) and
the model run with the eclipse. This phenomena is visi- the COSMO-DE including the solar eclipse (solid red
12 C. Köhler et al.: Impact of a solar eclipse on weather and photovoltaic production Meteorol. Z., PrePub Article, 2015
a) 20 March 2015
20000
PV-power [MW]
15000
14 GW
10000
7.5 GW
5000
0
b)
2500
2000
Reserve power [MW]
1500
1000
500
0
-500
-1000
-1500
c)
50,08
50,06
Frequency [Hz]
50,04
50,02
50,00
49,98
49,96
49,94
06:00 07:00 08:00 09:00 10:00 11:00 12:00 13:00 14:00 15:00
Time [CET]
Figure 17: Shown in a) is the projection of the PV-power feed-in, in b) the total in reserve power, and in c) the net frequency for Germany
on the 20 March 2015.
the power system, e.g. to increase the amount of reserve Germany due to the opposite cycles of the eclipse and
power. Months before the actual eclipse, sensitivity stud- the rising sun. The strongest influence was found at
ies accounting for the greatest uncertainty factor, namely 9:45 UTC under clear sky conditions. Considering all
cloud cover, were performed with the non-hydrostatic grid points over land in the simulations, the maximum
COSMO model of the DWD in order to assess the po- mean reduction in short wave radiation at the surface
tential maximal and minimal impact on solar radiation over Europe yields 31 % in comparison to non-eclipse
and temperature. These worst and best case NWP model conditions. For the smaller domain of Germany the de-
simulations were used to assess an estimation of possi- crease yields 68 %. Due to thermal inertia, the two me-
ble gradients in the German wide PV-power production ter temperature decline is delayed. The mean difference
during the solar eclipse. The gained experience helped over all land points is −2.67 °C at 10:00 UTC. A re-
to proactively prepare this special event. Days prior to duction in temperature remained perceptible for the rest
the solar eclipse these NWP model setups were used to of the day in the model runs. These NWP sensitivity
produce near real-time COSMO model forecasts. The studies were openly distributed (Köhler et al., 2014)
actual weather situation during the eclipse was similar and served as the basis for some of the many assess-
to the worst case scenario. Well planned precautions and ment studies dealing with PV power gradients on the
cooperation amongst the various participants guaranteed 20 March 2015, e.g., Kreifels et al. (2015); Köhler
that the power system stability was maintained through- et al. (2015). Also herein the NWP simulations were
out the event. used to estimate the German wide PV power produc-
Real case scenario simulations were conducted well tion. Due to the spatial nearly simultaneous and large
in advance using the non-hydrostatic COSMO models scale drop in irradiance, the simulations suggest strong
of the DWD. A clear sky scenario (20 March 2014) PV power feed-in gradients of up to 6.2 GW (15 min)−1
and an overcast scenario (16 March 2014), along with in the German power grid if clear sky conditions domi-
their reference model runs, were simulated in order to nate the day of the eclipse.
assess the worst and best case, respectively. Also, the During the eclipse actual near real-time NWP model
20 March 2014 was used for COSMO-EU simulations simulations were conducted and provided to the open
with artificially induced cloud free conditions. Therein public. On the 20 March 2015, high pressure influence
the total solar eclipse taking place in the morning hours caused clear sky conditions over most parts of Germany.
over northern Europe shows the highest impact over Only in the north-west (Amprion control area) persistent
14 C. Köhler et al.: Impact of a solar eclipse on weather and photovoltaic production Meteorol. Z., PrePub Article, 2015
Gueymard, C.A., 2004: The sun’s total and spectral irradiance Marty, J., F. Dalaudier, D. Ponceau, E. Blanc,
for solar energy applications and solar radiation models. – U. Munkhuu, 2013: Surface Pressure Fluctuations Pro-
Solar Energy 76, 423–453. duced by the Total Solar Eclipse of 1 August 2008. –
Helmert, J., B. Heinold, I. Tegen, O. Hellmuth, J. Atmos. Sci. 70, 809–823.
M. Wendisch, 2007: On the direct and semidirect ef- Morris, C., 2014: Photovoltaics: German grid braces for
fects of Saharan dust over Europe: A modeling study. – partial solar eclipse in 2015. – Renewables International
J. Geophys. Res. 112, D13. the Magazine, http://www.renewablesinternational.net/
Hess, P., H. Brezowsky, 1977: Katalog der Grosswetterla- german-grid-braces-for-partial-solar-eclipse-in-2015/150/
gen Europas 1881–1976. 3. verbesserte und ergänzte Aufl. – 452/79220/. Last accessed: 6 July 2015.
Berichte des Deutschen Wetterdienstes Nr. 113. Ratnam, V.M., M.S. Kumar, G. Basha, V.K. Anandan, A. Ja-
James, P.M., 2011: Summarizing ensemble forecast scenar- yaraman, 2010: Effect of the annular solar eclipse of 15 Jan-
ios using an objective classification of European synoptic uary 2010 on the lower atmospheric boundary layer over a
patterns. – EMS Annual Meeting, Berlin, Talk, http:// tropical rural station. – J. Atmos. Solar Terres. Phys. 72,
presentations.copernicus.org/EMS2011-499_presentation. 1393–1400.
pdf. Last accessed: 6 July 2015. Ritter, B., J.-F. Geleyn, 1992: A Comprehensive Radiation
Joe, P., M. Falla, 2004: Radar Visualization in the NinJo Scheme for Numerical Weather Prediction Models with Po-
Project. – Proceedings of the Third European Confer- tential Applications in Climate Simulations. – Mon. Wea. Rev.
ence on Radar in Meteorology and Hydrology (ERAD), 120, 303–325.
550–555. http://www.ninjo-workstation.com/fileadmin/files/ Schulz, J.-P., U. Schättler, 2011: Kurze Beschreibung
downloads/publications/Pub_ERAD04_P_550.pdf. Last ac- des Lokal-Modells Europa COSMO-EU (LME) und
cessed: 6 July 2015. seiner Datenbanken auf dem Datenserver des DWD. –
Kadygrov, E.N., E.A. Miller, A.V. Troitsky, 2013: Study Deutscher Wetterdienst, http://www.dwd.de/bvbw/generator/
of atmospheric boundary layer thermodynamics during total DWDWWW/Content/Forschung/FE1/Veroeffentlichungen/
solar eclipses. – IEEE Trans. Geosci. Remote Sensing 51, Download/LME__DBbeschr__1102,templateId=raw,
4672–4677. property=publicationFile.pdf/LME_DBbeschr_1102.pdf.
Kaldellis, J., P. Koepke, J. Reuder, J. Schween, 2014: Tem- Last accessed: 6 July 2015.
perature and wind speed impact on the efficiency of PV in- Standish, E.M., 1998: JPL Planetary and Lunar Ephemerides. –
stallations. Experience obtained from outdoor measurements Jet Propulsion Laboratory, Interoffice Memorandum 312.F-
in Greece. – Renewable Energy 66, 612–624. 98-048, ftp://ssd.jpl.nasa.gov/pub/eph/planets/ioms/de405.
Kazantzidis, A., A.F. Bais, C. Emde, S. Kazadzis, C.S. Zere- iom.pdf. Last accessed: 6 July 2015.
fos, 2007: Attenuation of global ultraviolet and visible irra- Subrahamanyam, D.B., T.J. Anurose, 2011: Solar eclipse in-
diance over Greece during the total solar eclipse of 29 March duced impacts on sea/land breeze circulation over Thumba: A
2006. – Atmos. Chem. Phys. 7, 5959–5969. case study. – J. Atmos. Solar Terres. Phys. 73, 703–708.
Koepke, P., J. Reuder, J. Schween, 2001: Spectral variation Subrahamanyam, D.B., T.J. Anurose, M. Mohan, M. San-
of the solar radiation during an eclipse. – Meteorol. Z. 10, tosh, N.V.P.K. Kumar, S. Sijikumar, S.S. Prijith, M. Aloy-
179–186. sius, 2011: Atmospheric Surface-Layer Response to the An-
Köhler, C., A. Steiner, D. Lee, J. Thieler, B. Ritter, 2014: nular Solar Eclipse of 15 January 2010 over Thiruvananthapu-
Case studies of the effects of a total solar eclipse on weather ram, India. – Bound.-Layer Meteor. 141, 325–332.
with varying degrees of cloud cover using a local numerical Vogel, B., M. Baldauf, F. Fiedler, 2001: The Influence of
weather prediction model. – PANGAEA – Data Publisher for a solar eclipse on temperature and wind in the Upper-Rhine
Earth & Environment Service, http://doi.pangaea.de/10.1594/ Valley - A numerical case study. – Meteorol. Z. 10, 207–214.
PANGAEA.839163. Last accessed: 6 July 2015. Weinhold, N., 2014: Wenn der Mond mitmischt – Forscher
Köhler, C., H. Ruf, A. Steiner, D. Lee, J. Thieler, others, untersuchen Einfluss einer Sonnenfinsternis auf Versorgung. –
2015: Nutzung Numerischer Wettervorhersagen in der Sim- Erneuerbare Energien das Magazin 12.
ulation von Verteilnetzen: Die Effekte einer Sonnenfinsternis Weniger, J., J. Bergner, T. Tjaden, V. Quaschning,
auf netzgekoppelte PV-Anlagen und Netztransformatoren. – 2014: Einfluss der Sonnenfinsternis im März 2015 auf
Tagungsband OTTI, Ed, 30. Symposium Photovoltaischer So- die Solarstromerzeugung in Deutschland. – Study, http://
larenergie, Kloster Banz, Bad Staffelstein 1st ed. Regensburg, pvspeicher.htw-berlin.de/wp-content/uploads/2014/10/
2015, ISBN: 9783943891454. HTW-Berlin-Studie-Einfluss-der-Sonnenfinsternis-im-M&
Kreifels, N., S. Killinger, J.N. Mayer, B. Müller, auml;rz-2015-auf-die-Solarstromerzeugung-in-Deutschland.
C. Wittwer, others, 2015: Sonnenfinsternis am pdf. Last accessed: 6 July 2015.
20. März 2015 – Auswirkungen auf die System- Zerefos, C., N. Mihalopoulos, P. Monks (Eds), 2007: The
stabilität der deutschen Stromversorgung. – Fraun- total solar eclipse of 2006 and its effects on the environment. –
hofer ISE Bericht, http://www.ise.fraunhofer.de/de/ Atmos. Chem. Phys., Special Issue, published online, http://
veroeffentlichungen/veroeffentlichungen-pdf-dateien/ www.atmos-chem-phys.net/special_issue87.html.
studien-und-konzeptpapiere/studie_sonnenfinsternis.pdf.
Last accessed: 6 July 2015.