Atmospheric Research 100 (2011) 310–317

Contents lists available at ScienceDirect

Atmospheric Research

j ourna l homepage: www.e lsev ie r.com/ locate /atmos

A climatology of cloud-to-ground lightning over Estonia, 2005–2009

S.E. Enno⁎Department of Geography, University of Tartu, Vanemuise 46, 51014 Tartu, Estonia

a r t i c l e i n f o

⁎ Tel.: +3727375824; fax: +3727375825.E-mail address: sven-erik.enno@ut.ee.

0169-8095/$ – see front matter © 2010 Elsevier B.V.doi:10.1016/j.atmosres.2010.08.024

a b s t r a c t

Article history:Received 12 January 2010Received in revised form 20 August 2010Accepted 27 August 2010

This paper presents the spatial and temporal distribution of cloud-to-ground lightning overEstonia and the adjacent sea for the period 2005–2009. Data collected by the NORDLIS (NORDicLightning Information System) lightning detection network was used. The spatial distributionof lightning was calculated in a 10×10 km grid. A total of 172,613 cloud-to-ground flasheswere registered in the area under observation during 2005–2009. The annual mean cloud-to-ground flash density over the area was 0.34 flashes km−2 year−1. The lowest values foundwere less than 0.10 flashes km−2 year−1 and the highest flash densities were 0.80–1.01 flasheskm−2 year−1. The monthly distribution of lightning showed the highest activity in July andAugust. Of all the registered flashes, 99.4% were reported from May to October. The dailydistribution of flashes showed single days on which thunderstorm activity was very high,against a background of much lower everyday activity. The diurnal distribution of lightningshowed an evident peak between 15:00 and 17:00 local time over land. Over the sea, a flattermaximum lay between 13:00 and 21:00. Minimum lightning activity occurred between 22:00and 06:00 over land and from 02:00 to 09:00 over the sea. Our work revealed that the spatialand temporal characteristics of cloud-to-ground lightning over Estonia generally resemble thecharacteristics found at other mid-latitude study sites.

© 2010 Elsevier B.V. All rights reserved.

Keywords:DetectionCloud-to-ground lightningFlash densityEstonia

1. Introduction

Thunderstorms are among the most damaging weatherevents. Thus, it is useful to know the spatial and temporalcharacteristics of the occurrence of thunderstorms andlightning. Nowadays, a variety of data sources are availablefor use in studying the spatial and temporal distribution ofthunderstorms. Visual observations at meteorological sta-tions are the oldest available records. At many locations, thedata for annual and monthly numbers of thunderstorm daysgo back further than 100 years, thereby making long-termclimatic studies possible. For example, Changnon andChangnon (2001) used data of 86 stations for the period1896–1995 to study long-term fluctuations in the annualnumbers of thunderstorm days in the contiguous UnitedStates. However, it has been demonstrated that the intensityof thunderstorms, as well as exterior factors, such as the

All rights reserved.

background noise and daylight, significantly affect visualobservations (Reap and Orville, 1990). Hence, it is wise to usedata sets that are more reliable.

Lightning detection networks allow the collection ofcontinuous lightning data that has high spatio-temporalaccuracy and enable discrimination between cloud-to-groundand cloud flashes. The data is processed and made available inreal time. Cummins et al. (1998) describe in detail the differentlightning sensors and properties of lightning detection net-works. The development of national and international lightningdetection systems over the last 20 to 30 years has made itpossible to study thunderstorm climates by using flash densityas the main measure. Flash density is the average number of(usually cloud-to-ground) lightning flashes per area unit pertime unit. The measure that is used most widely is the annualnumber offlashes per square kilometre. It ismore accurate thanthe traditionally used average annual number of thunderstormdays, because it not only reflects the presence of thunder-storms, but also gives an overview of the number and spatialdistribution of lightning flashes.

311S.E. Enno / Atmospheric Research 100 (2011) 310–317

Diurnal (hour-to-hour), seasonal (day to day ormonth-to-month) and annual (year-to-year) variations in lightningactivity are widely used for temporal analyses (Tuomi andMäkelä, 2008a). It is also possible to calculate the annualnumber of thunderstorm days or thunderstorm hours fromthe lightning detection data. Huffines and Orville (1999) didso for the USA, while Rivas Soriano and De Pablo (2002b)made similar calculations for the Iberian Peninsula.

Many reports of regional studies on the spatial andtemporal distribution of lightning have been published. Themost widely used spatial grid spacing for calculating flashdensities seems to be 0.2°×0.2°. This spacing has been usedfor the USA (Orville and Huffines, 2001), the USA and Canada(Orville et al., 2002), the Iberian Peninsula (Rivas Sorianoet al., 2005) and Sweden (Sonnadara et al., 2006). However,the areas of the cells in this kind of grid are unequal; theydecrease towards high latitudes. Many authors prefer equal-area grid cells. Burrows et al. (2002) used a 20×20 km gridfor Canada, and Antonescu and Burcea (2010) used the samegrid spacing for Romania. Tuomi and Mäkelä (2008a) used a10×10 km spatial grid to calculate flash densities whenanalyzing the thunderstorm climate of Finland. Schulz et al.(2005) used 10×10 and 1×1 km grids for geographical plotswhen studying the cloud-to-ground lightning over Austriafrom 1992 to 2001. The 1×1 km grids were too small, becauseindividual radio towers and mountain summits create noisethat clearly affect the results.

Relationships between lightning activity and other atmo-spheric factors or synoptic situations have also been exam-ined extensively in modern lightning climatology. Tuomi andMäkelä (2008a) studied the relationships between synopticweather types and lightning activity. A similar analysis wasmade for the Iberian Peninsula by Clemente et al. (2004).Tuomi and Larjavaara (2005) used lightning detectionnetwork data to identify and analyze flash cells in thunder-storms. Characteristics of cloud-to-ground flashes over theIberian Peninsula were compared with geographical latitudeand longitude (Rivas Soriano et al., 2002), and with the seasurface temperature (De Pablo and Rivas Soriano, 2002). Theeffect of urban pollution on cloud-to-ground lightning

Fig. 1. a) Location of the NORDLIS lightning detection network (black rectangle). b)(black rectangle).

activity was analyzed for the Midwestern USA (Westcott,1995) and central Spain (Rivas Soriano and De Pablo, 2002a).

The first weather and thunderstorm records in the archiveof the Estonian Meteorological and Hydrological Institute(EMHI) date back to the 18th century, but regular observa-tions with continuous data rows were started in the secondhalf of the 19th century. These records also include theoccurrence of thunderstorms. The annual numbers of thun-derstorm days for the entire 20th century are available atsome weather stations. However, until 2005, the only type ofthunderstorm data that was collected in Estonia was visualrecords. Many local-scale studies have been written on thebasis of visual thunderstorm observations. It is known that,on average, there are 15–25 thunderstorm days per year inEstonia. Unfortunately, none of the Estonian thunderstormstudies have so far been published internationally.

The study reported herein yielded the first results for thespatial and temporal distribution of cloud-to-ground light-ning over the Estonian landmass and the adjacent sea,including islands. Estonia was incorporated into the NORDLISlightning detection network at the end of 2004. Thus, in theinterests of using the most reliable data, data from 2005 to2009 were used.

The remainder of the paper is organized as follows. InSection 2, the data and methods are described. In Section 3,the results are presented. In Section 4, the results arediscussed. Section 5 concludes.

2. Data and methods

The study reported herein used data from the NORDLIS,which incorporates lightning sensors in Norway, Sweden,Finland, and Estonia. The central unit of the system operatesin Finland and belongs to the Finnish Meteorological Institute(FMI). Fourteen sensors in Norway, nine in Sweden, seven inFinland, and one in Estonia are connected to the central unit(Fig. 1). The system uses mainly IMPACT sensors, with someof its successor model, the LS7000, also deployed. Thesesensors use low-frequency electromagnetic radiation todetect lightning. NORDLIS is mainly capable of locating

Locations of the NORDLIS sensors in 2009 (black circles) and our study area

0

2

4

6

8

10

12

-100 -80 -60 -40 -20 0 20 40 60 80 100

Peak current kA

Nu

mb

er o

f fl

ash

es (

tho

usa

nd

s)

Fig. 2. Peak current distribution of registered cloud-to-ground flashes in thestudy area during the period 2005–2009.

312 S.E. Enno / Atmospheric Research 100 (2011) 310–317

cloud-to-ground lightning, because intracloud lightningemits VLF/LF radiation mainly in the range of smalleramplitudes and with waveforms that do not pass the applieddiscrimination filter. Tuomi (2005) and Cummins et al. (1998)describe the properties of IMPACT sensors in greater detail.

The lightning detector was installed in Estonia in the end of2004, so from2005, the efficiency of the networkwas increasedover Estonia. The initial dataset thatweusedwasobtained fromthe EMHI; it consists of 228,997 flashes registered during thefive-year period 2005–2009.Allflashes typed as cloud lightningwere removed from the dataset, to ensure that, as far as ispossible, our final dataset contained only records of cloud-to-ground lightning. However, it should be noted that, especiallyfor stroke currents below 15 kA, the system cannot discrimi-nate reliably between intracloud and cloud-to-ground strokes.

A 10×10 km spatial grid was used to calculate flashdensities. Although Schultz et al. (2005) demonstrate thatlarger grids give more accurate results, we decided to use a10×10 km grid, because the study area is quite small, only410×250 km. The same grid size worked well for theNORDLIS data analysis in Finland 1998–2007 (Tuomi andMäkelä, 2008a). Yearly lightning maps with this grid spacingwere also published by Tuomi and Mäkelä (2006), Tuomi andMäkelä (2007), and Tuomi and Mäkelä (2008a,b,c). The gridthat we used is based on the Estonian coordinate system L-EST97, which makes it possible to use equal-sized squarecells. The grid covered 102,500 km2 between 57.5° and 59.8°N and 21.0° and 28.5° E. The study area encompasses Estonia,the eastern part of the Baltic Sea, the southern part of the Gulfof Finland, Northern Latvia, and the western edge of Russia(Fig. 1). A total of 172,613 cloud-to-ground flashes wereregistered in the study area in the period 2005–2009.

All the data was imported to Idrisi 32 software, where thespatial grid was created and all calculations of flash densitywere done. The results of the calculations were imported toArcGIS, where the final maps were designed.

The number of ground flashes per square kilometre wasused as the main characteristic of flash density. Flash densitywas analyzed for the whole five-year period, and year by yearand month by month. No corrections for detection efficiencyweremade in this analysis. The first-stroke detection efficiencyof the NORDLIS network in Finland is estimated to range from88% to over 99% for cloud-to-ground lightning (Tuomi andMäkelä, 2008c). Lowest efficiency was found for Lapland nearthe northern edge of the NORDLIS network, but highestefficiency was found in Southern Finland, and Estonia is only100–400 km from Southern Finland. This would indicate thatprovided network coverage is similar, we could expect levelsof efficiency similar to that achieved in Southern Finland. Onthe other hand, no NORDLIS sensors have been installed in thesouth and east of Estonia. As a consequence, the efficiencywith which cloud-to-ground flashes are detected in the studyarea probably decreases slightly from the north to the southand the west to the east.

In 2008, one sensor was removed from Western Finland,because it was broken. Accordingly, it is possible that a smalldrop in detection efficiency has occurred since 2008.However, this possibility is not significant for our study,because the four sensors on the southern coast of Finland thatare closest to Estonia were working throughout the wholestudy period.

It is possible to estimate approximately the actuallyrelevant detection efficiency over a given area by knowingthe average peak currents of observed flashes as well as thepeak current distribution of flashes (Pinto, 2008). For thestudy area, during 2005–2009 the average peak currentswere17.7 and 19.9 kA for negative and positive flashes, respec-tively. The current distribution of registered flashes ispresented in Fig. 2.

In addition to the spatial distribution, the temporaldistribution of lightning was studied month by month andday by day. Diurnal variations in the occurrence of flasheswere analyzed for the whole study area and also separatelyfor land and the sea. All flashes over the islands were added tothe sea data and all flashes over inland lakes were counted asland data. As justification for this measure, we determinedthat the diurnal lightning distribution over the biggest island,Saaremaa (area 2700 km2), was similar to the sea areas, andover the biggest lake, Lake Peipsi (area 3500 km2), itresembled that of the land areas.

3. Results

The annual average flash density for the whole study areafor the period 2005–2009 was found to be 0.34 flashes km−2

year−1. The interannual variability is remarkable. In 2006,only 17,000 flashes were registered (flash density 0.17 flasheskm−2 year−1). In 2008, 23,500 and in 2005 about 25,000flashes were registered over the study area (flash densities0.23 and 0.24 flashes km−2 year−1, respectively). Manymorethunderstorms occurred in 2007 and 2009. In 2009, 50,000flashes with an average density of 0.49 flashes km−2 year−1

were registered. In 2007, more than 57,000 recorded flashesgave an average density of 0.56 flashes km−2 year−1.

The spatial distribution of ground flashes over Estoniaduring the period 2005–2009 (Fig. 3) is rather complicated. Thesea areas show lower densities than the land areas. The lowestlightning frequencies (less than 0.1flashes km−2 year−1)wereobserved at the western edge of the study area over the openwaters of the Baltic Sea. Most sea areas showed values lessthan 0.4 flashes km−2 year−1 and no more than 0.5 flasheskm−2 year−1 were registered on the open sea and the islands.

Fig. 3. Annual average flash density (flashes km−2 year−1) in the study area during the period 2005–2009.

313S.E. Enno / Atmospheric Research 100 (2011) 310–317

However, the eastern part of the largest island Saaremaashowed a peak with 0.61 flashes km−2 year−1. Land areasand also some coastal districts exhibit higher flash densities.On land, 0.3–0.6 flashes km−2 year−1 is common and theflash density generally increases from the southwest to thenortheast. Maxima with 0.6–0.8, in some places 0.8–1.01,flashes km−2 year−1 are mostly located in the northeasternpart of the Estonian land area. The only exception is a smallarea with 0.6–0.9 flashes km−2 year−1 on the southwesterncoast. Another interesting feature is the fact that high flashdensities on the northeastern land areas continue over thewaters of the Gulf of Narva towards the Russian KurgolovoPeninsula (the northeastern corner of the study area).

The monthly distribution of registered flashes (Fig. 4)shows the main thunderstorm season from May to August.The most active months of July and August both give about28% of all registered flashes, so more than half of all theflashes during the period 2005–2009 occurred in these2 months. Roughly 21% of the flashes occurred in May and14% in June. More than 90% of all the flashes occurred fromMay to August. About 4% of the annual number of flashesoccurred in each of September and October, 0.4% in April, and0.2% in November. Almost no thunderstorms occurred from

0

10

20

30

40

50

60

I II III IV V VI VII VIII IX X XI XII

Month

Nu

mb

er o

f fl

ash

es (

tho

usa

nd

s)

Fig. 4. Monthly distribution of flashes during the period 2005–2009.

December to March. In December and January, only eightflashes were registered over the whole study area during the5-year period. During the period 2005–2009, nine flashesoccurred in March and 43 in February.

Given that almost no thunderstorms were recordedbetween November and March, the average daily numbersof flashes over the study area (Fig. 5) are given from April toOctober. For each date, the number in the figure denotes afive-year average. It can be seen that the main thunderstormseason begins in mid-May and lasts until the beginning ofSeptember. However, some events occurred outside of themain season. These include some days with fairly highthunderstorm activity at the end of September and at thebeginning of October. Nevertheless, it is obvious thatthunderstorms are not frequent in April, September, andOctober. The main season is characterized by an average dailyactivity of about 100–200 flashes over the study area. Inaddition, there are a lot of individual peaks with 500–1500flashes per date. Two extremely high peaks are visible. 13June has an average activity of 2700 flashes per day and 19July has 2300 flashes per day. Every strong peak can be

0.0

0.5

1.0

1.5

2.0

2.5

3.0

Date

Ave

rag

e n

um

ber

of

flas

hes

(th

ou

san

ds)

APR MAY JUN JUL AUG SEPT OCT

Fig. 5. Daily average numbers of registered flashes (in thousands) from Aprilto October over the study area during the period 2005–2009.

314 S.E. Enno / Atmospheric Research 100 (2011) 310–317

associated with a single heavy storm that occasionallyoccurred on that date in one of the years studied. Thisinfluence of a single heavy storm to the lightning statistics isdiscussed further in Section 4, with some examples.

The diurnal distribution of lightning (Fig. 6) differsremarkably in the land and sea areas, so both distributionsare given. Over land, an evident peak occurred between 15:00and 17:00 local time. Each of the hours at which the peakoccurs contains more than 10% of all registered flashes. It maybe concluded that one third of all registered ground flashesover land occurred between 15:00 and 17:00. During theevening hours, there is a rapid decrease in lightning activity.The minimum activity occurred between 22:00 and 06:00,when each hour contained only 1–2% of all registered flashes.Hence, the lightning activity over land during the night-timeminimum period is roughly 10 times lower than during thedaytime maximum period.

Diurnal variation is much smaller over the sea. Themaximum is flat and lies between 13:00 and 21:00 with anevident peak at 14:00. It is quite difficult to determine the endof the maximum, because activity continues to be quite highuntil 01:00. Theminimumoccurs from 02:00 to 09:00. Each ofthe hours during which maximum activity occurs contains 6–8% and each of the minimum hours 2–3% of all registeredflashes over the sea. So, lightning activity over the sea duringthe night and morning minimum is about 2–3 times lowerthan during the day and evening maximum.

4. Discussion

The average annual flash density over the whole studyarea was 0.34 flashes km−2 year−1 and varied from less than0.10 flashes km−2 year−1 to 1.01 flashes km−2 year−1. Thesevalues are low on a global scale. Similar lightning detectionnetworks register up to 6–8 flashes km−2 year−1 in north-eastern Italy (Bernardi and Ferrari, 2004) and over 9 flasheskm−2 year−1 in Florida, USA (Orville and Huffines, 2001). So,our flash densities are about 20–30 times smaller than thehighest measured densities in Europe and the USA.

The same average flash density of 0.34flashes km−2 year−1

was found for Finland during 1998–2007 by Tuomi andMäkelä (2008a). It should be taken into account that theFinnish study area does not include open sea low-activity

Fig. 6. Diurnal distribution of flashes over land (left) a

areas, whereas our study area does. It is also worthmentioning that the years 1998 to 2002, which probablyexperienced much greater thunderstorm activity, wereincluded by Tuomi and Mäkelä (2008a), but they are notrepresented in our dataset. Years 2005–2008 exhibited flashdensities below 0.2 flashes km−2 year−1 over the Finnishstudy area, which is well below the 10-year average, as wellas below the Estonian values for the same years. So it can beconcluded that thunderstorm activity over Estonia is higherthan over Finland. It is also obvious that Sweden experiencesfewer flashes than Estonia. During the period 1987–2000, theaverage annual flash density over Sweden varied from 0.03flashes km−2 year−1 in the north to 0.4 flashes km−2 year−1

near the southern coast of the country (Sonnadara et al.,2006). So, the most active locations in southern Swedenexperience as many flashes as the average in the Estonianstudy area. The annual numbers of registered flashes vary upto five times between individual years during the 13-yearperiod studied by Sonnadara et al. (2006). Similar high year-to-year variations are also characteristic for Estonia. Groundflash density comparable with Estonia was also found byAntonescu and Burcea (2010) for eastern and northwesternRomania. In North America, flash densities similar to Estoniacan be found in northern andwestern parts of Canada (Orvilleet al., 2002) and in western and northeastern areas of the USA(Orville and Huffines, 2001).

The spatial distribution of flashes (Fig. 3) is difficult toexplain in terms of the main topographic features, such asuplands. The 5-year study period is probably too short andthe uplands are too low. The Finnish 10-year study perioddoes not show a correlation between flash density andtopography either (Tuomi and Mäkelä, 2008a), so an evenlonger period would be needed to establish one. However,our study shows that open waters of the Baltic Sea in thewestern edge of the study area have the lowest flashdensities. The water in this region is a relatively cool surfaceduring summertime with little or almost no spatial variationsin temperature; hence it is not conducive to convection andthunderstorms. In contrast, the Gulf of Finland, which liesbetween land areas of Estonia and Finland, does not inhibitthunderstorms as much as the open sea. Very high flashdensities are observed over the eastern part of the gulf.However, these storms are probably not initiated by the sea,

nd over sea (right). 14 h local time is 12 h UTC.

315S.E. Enno / Atmospheric Research 100 (2011) 310–317

but formed over northeastern Estonia. Due to the fact thatwinds from the south and southwest are common, thesestorms often move to the gulf and frequently continue toproduce many flashes over the water.

Northeastern Estonia seems to be conducive to thedevelopment of thunderstorms. There are large swampsthat can supply warm humid air, as well as uplands that canforce updrafts. A previous, visual observations-based localstudy indicated that the highest annual numbers of thunder-storm days occur in the northeastern part of the country.However, the cause of the prevalence of thunderstorms inthis area is not immediately evident. Other upland areas insouthern and southeastern Estonia do not show lightningmaxima, even though they are higher than the northeasternuplands, which would indicate that the prevalence ofthunderstorms in northeastern Estonia is not due to thepresence of uplands. In addition, one maximum lies just onthe southwest coast, on coastal lowland, which wouldconfirm that the prevalence of thunderstorms is not associ-ated with the presence of uplands. However, this maximum isprobably associated with one very powerful storm in thatregion on 19 July 2009, and thus it is not a long-term climaticfeature. It is evident that the available data is not currentlysufficient to account for the prevalence of thunderstorms overnortheastern Estonia. Further data is needed. It is to be hopedthat such further data will minimize the influence of singlestrong thunderstorm events and enable us to improve ourmap of the distribution of flashes according to climate. Longerdata rows will also mean more flashes per grid cells andhence a higher statistical accuracy for values for the density ofground flashes (Diendorfer, 2008).

The monthly distribution of lightning (Fig. 4) is not verytypical, but its main properties are still similar to those foundin other mid-latitude studies.

July was found to be the most active month. The same wasfound for Finland (Tuomi and Mäkelä, 2008a), Sweden(Sonnadara et al., 2006), Austria (Schulz et al., 2005), Canada,(Burrows et al., 2002) and the USA (Orville and Huffines,2001). July is the warmest month in mid-latitude continentalareas, so the development of a greater number of activethunderstorms in this month is rendered likely by the hightemperatures.

Other mid-latitude studies found that June and Augusthave clearly lower activities than July, and that inMay activityis markedly lower than June and August. These results are, atfirst sight, in contrast to those found in our study. We foundthat August is almost as active as July and that May has amarkedly higher number of registered flashes than June.However, these initial discrepancies can be explained away.

The high activity that we found in August may beattributed, at least in part, to the influence of the sea. Seawater attains its highest temperature in August, which hasbeen recorded to be the warmest month at some areas of thecoast in Estonia. Hence, the highest lightning activity over thesea is expected to occur not in July, but in August, andthe other studies did not cover open sea areas.

The high lightning activity that we found in May seems tobe occasional; it mainly comprises unusually powerful andwide storms during the last days ofMay 2007. Hence, it wouldbe unwise to take the results for May as being representativeof activity in May in general. In contrast with May, June

showed unusually low lightning activity during the period2005–2009. For the same period, unusually low activity inJune is also reported for Finland by Tuomi and Mäkelä(2008a). If the high activity during the last days of May 2007and the low activity during June is taken to be an anomaly,our results will be in agreement with those of other studies.

Of all the flashes that were registered during the studyperiod, 95.6% occurred from May to September and 99.4%from May to October. These results are very similar to resultsobtained for Austria, Romania, and Canada. Schulz et al.(2005) reported that 96% of flashes in Austria during theperiod 1992–2001 occurred from May to September. Anto-nescu, and Burcea (2010) found that about 98% of allregistered lightning in Romania during the period 2003–2005 and in 2007 was detected from May to September. InWestern Canada, 98.9% and in Eastern Canada, 93.0% offlashes recorded from 1998 to 2000 occurred in the 6 warmmonths from May to October (Burrows et al., 2002). FromNovember to March, almost no thunderstorms occurred inEstonia. The mean temperatures for these months are belowzero, so vertical air movements do not occur and as a result,thunderclouds do not form. Only occasional flashes producedby powerful cold fronts are possible.

The daily distribution of flashes (Fig. 5) shows that themain thunderstorm season runs from the middle of May toearly September. During this period, 100–200 flashes per dayoccur over the study area. Such activity is characteristic oflocal storms. The flashes are often clustered in small parts ofthe study area, whereas no thunderstorms occur in the rest ofthe area. Frontal storms produce much more flashes, becausethey cover much larger areas, sometimes affecting the wholestudy area. As seen in Fig. 5, they cause individual strongpeaks even in the 5-year average for daily activity. Most daysthat average more than 1000 flashes over the study area areassociated with the occurrence of individual heavy frontalstorms in one of the years from 2005 to 2009. Dramaticexamples are 13 June and 19 July 2009, on which 13,000 and9500 flashes occurred, respectively. These two days togetherproduced about 45% of the total ground flashes registered in2009. All the increased activity at the end of May is due topowerful frontal storms that occurred from 26 to 31 May2007. There was another outbreak of heavy storms from 21 to24 August 2007. One late heavy storm, which is easilyrecognizable in Fig. 5, occurred on 1 October 2006. Within afew hours, it produced about one third of the total number offlashes that were registered in 2006. Thus, in Estonia, most ofthe annual lightning activity seems to be concentrated in afew very active days. Similar results are reported for Canada.In southern Ontario, the 4 most active days in 1998 and 1999accounted for more than 25% of annual lightning. In someplaces, about half of the annual lightning in 1998 occurred ina 26-hour period and 25% in just 4 h (Burrows et al., 2002).

All strong storms in the Estonian study area that occurredin the period from 2005 to 2009 are associated with fronts. Ina usual situation, the advection of warm and humid air fromthe south is followed by a cold front from the west, producingpowerful and widespread thunderstorms. Due to the fact thatthe study area is quite small, it takes less than 24 h for mostfronts to cross it. So on one day, an activity peak can appear inthe lightning data, while there is no or very weak activityduring the previous and next days. Only stationary or very

316 S.E. Enno / Atmospheric Research 100 (2011) 310–317

slow-moving fronts can generate very active periods that lastfor 2–5 days, as happened in May and August 2007.

The daily distribution of lightning was also published forFinland by Tuomi and Mäkelä (2008a). The report was basedon 10-year data, so the results are more homogenous. Manyindividual peaks are still visible, but these do not stand out asmuch as the peaks that we found for Estonia for 2005 to 2009.We expect that over a longer timescale, we would find amoreeven daily distribution of lightning over Estonia, in which theeffect of individual storms would be lessened.

Overall, individual powerful storms do not have a strongeffect on the overall 5-year spatial distribution of flashdensity. They tend to produce more lightning over land andin northeastern Estonia, and less over the coastal and open-water areas. In so acting, they act as local storms. The onlyexception in our study area during the study period is theheavy storm on 19 July 2009. On that afternoon, an unusuallyhigh number of flashes were produced at some locations onthe southwest coast by a strong frontal storm. As mentionedearlier, the high activity on the southwestern coastal area(Fig. 3) was caused by that storm. The storm peaked ataround 14:00 local time and a lot of flashes also hit the sea.The single peak that is visible at 14:00 in the diurnaldistribution of flashes over the sea (Fig. 6) was also causedby that storm.

The diurnal distribution of flashes (Fig. 6) is much moreeven over the sea. This phenomenon is probably associatedwith the fact that the sea surface temperature varies little ornot at all between day and night. The distribution of landflashes correlates well with the diurnal temperature cycleover land. A strong peak occurred between 15:00 and 17:00local time, just after the usual daily maximum temperatures.This peak time corresponds well with the results of othersimilar studies conducted in Finland (Tuomi and Mäkelä,2008a), Austria (Schulz et al., 2005), the Iberian Peninsula(Rivas Soriano et al., 2005), Romania (Antonescu and Burcea,2010), and Canada (Burrows et al., 2002). Minimum activityis usually found to occur during the night and in the earlymorning. This is also the case in the Estonian land area.

The diurnal distribution of lightning over the sea inFinland was studied by Tuomi and Mäkelä (2008a). Theirresults are quite similar to ours. The maximum is flat andelongated towards the evening hours. However, at least thefirst half of the maximum over the sea occurred at the sametime the land maximum. It is evident that some stormsdevelop over land, but then drift to the sea, thus it is notsurprising that during the land maximum hours, the seaareas, especially near the coast, also experience a highnumber of flashes. As the water maintains its temperature,the dissipation of storms during evening hours is slower. Thatcan explain why the diurnal maximum is elongated towardsnight over the sea.

5. Conclusions

We have provided an overview of the spatio-temporaldistribution of cloud-to-ground lightning over Estonia andthe adjacent sea during the period 2005–2009. The total studyarea covers 102,500 km2 in Estonia, the eastern part of theBaltic Sea, the southern part of the Gulf of Finland, Northern

Latvia, and the western edge of Russia. Flash densities werecalculated for a 10×10 km spatial grid.

A total of 172,613 cloud-to-ground flashes were registeredin the study area during the study period. About three timesas many flashes occurred during the high-activity years of2007 and 2009 than in 2006.

The annual mean cloud-to-ground flash density over thearea was 0.34 flashes km−2 year−1. The lowest values ofless than 0.10 flashes km−2 year−1 were found over theopen waters of the Baltic Sea in the western edge of thestudy area. Flash densities over the sea were generallylower than over land. The only exception was the easternpart of the Gulf of Finland, where remarkably high densitieswere found near the northeast coast of Estonia. The highestflash densities of 0.80–1.01 flashes km−2 year−1 wereconcentrated in northeastern Estonia. All these values arelow on a global scale and are comparable with flashdensities in Finland, Sweden, western USA, and northernand western Canada. No clear correlation between land-forms and lightning activity appeared. Study of data thatcovers longer periods is required to specify the climaticdistribution of flashes in more detail and demonstrate thestatistical accuracy of results.

The mid-latitude specific annual distribution showed99.4% of annual flashes occurring from May to October and95.6% occurring from May to September. July and Augustwere themost active months. Together, about 56% of the totalannual number of flashes occurred during these months. Mayshowed high and June unusually low activity. Almost nothunderstorms occurred from November to March.

The average daily numbers of flashes over the study areashow the main thunderstorm season running from themiddle of May to early September. However, some days onwhich activity was quite high lay outside the main season, atthe end of September and at the beginning of October. Themain season is characterized by an average daily activity ofabout 100–200 flashes over the study area. During this period,there are high peaks of activity that are associatedwith strongand wide frontal storms. Up to 13,000 flashes have beenregistered during one day. Just a few days can yieldmore thanhalf of the total number of flashes registered annually. In atypical scenario that is characterized by high activity,advection of warm and humid air from the south is followedby a cold front from the west, which produces powerfulthunderstorms that cover a large area.

The diurnal distribution of lightning differs markedlybetween the land and the sea areas. Over land, an evidentpeak occurred between 15:00 and 17:00 local time. One thirdof all registered flashes over land occurred from 15:00 to17:00. Minimum lightning activity occurred between 22:00and 06:00. These times correspond well with the results ofother similar studies conducted on mid-latitude countries.Lightning activity over land during the night-time minimumwas roughly 10 times lower than during the daytimemaximum. Diurnal variations were much smaller over thesea. The maximum is flat and lies between 13:00 and 21:00.Theminimum occurs from 02:00 to 09:00. Due to the fact thatthe water maintains its temperature at night, the dissipationof storms during the evening hours is slower over the sea.This dissipation might explain why the diurnal maximum iselongated towards night over the sea.

317S.E. Enno / Atmospheric Research 100 (2011) 310–317

Acknowledgements

The author would like to thank Tapio J. Tuomi and AnttiMäkelä from the Finnish Meteorological Institute for helpfulmaterials and advice. The author also would like to thank Dr.David M. Schultz and two anonymous reviewers for theirhelpful comments. This research was supported by theEstonian Science Foundation grant No. 7510.

References

Antonescu, B., Burcea, S., 2010. A cloud-to-ground lightning climatology forRomania. Mon. Weather Rev. 138, 579–591.

Bernardi, M., Ferrari, D., 2004. Evaluation of the LLS efficiency effects on theground flash density, using the Italian lightning detection system SIRF. J.Electrostat. 60, 131–140.

Burrows,W.R., King, P., Lewis, P.J., Kochtubajda, B., Snyder, B., Turcotte, V., 2002.Lightningoccurrencepatterns over CanadaandadjacentUnitedStates fromlightning detection network observations. Atmos. Ocean 40, 59–81.

Changnon, S.A., Changnon, D., 2001. Long-term fluctuations in thunderstormactivity in the United States. Clim. Change 50, 489–503.

Clemente, T., De Pablo, F., Rivas Soriano, L., 2004. Circulation weather typesand cloud-to-ground flash density over the Iberian Peninsula. Int. J.Climatol. 24, 109–12310.1002/joc.917.

Cummins, K.L., Krider, E.P., Malone, M.D., 1998. The U.S. National LightningDetection Network and applications of cloud-to-ground lightning databy electric power utilities. IEEE Trans. Electromagn. Compat. 40,465–480.

De Pablo, F., Rivas Soriano, L., 2002. Relationship between cloud-to-groundlightning flashes over the Iberian Peninsula and sea surface temperature.Q. J. R. Meteorol. Soc. 128, 173–183.

Diendorfer, G., 2008: Some comments on the achievable accuracy of localground flash density values. Proceedings of the 29th InternationalConference on Lightning Protection (ICLP 2008), Uppsala, Sweden 23–26June 2008. (available at http://www.aldis.at/pdf/12_ICLP_2008_DIN.pdf).

Huffines, G.R., Orville, R.E., 1999. Lightning ground flash density andthunderstorm duration in the continental United States: 1989–96. J.Appl. Meteorol. 38, 1013–1019.

Orville, R.E., Huffines, G.R., 2001. Cloud-to-ground lightning in the UnitedStates: NLDN results in the first decade, 1989–98. Mon. Weather Rev.129, 1179–1193.

Orville, R.E., Huffines, G.R., Burrows, W.R., Holle, R.L., Cummins, K.L., 2002.The North American Lightning Detection Network (NALDN)—firstresults: 1998–2000. Mon. Weather Rev. 130, 2098–2109.

Pinto, O., 2008: An overview of cloud-to-ground lightning research in Brazilin the last two decades. Proceedings of the 20th International LightningDetection Conference, Tuscon, Arizona, USA, 21–23 April 2008. (availableat http://www.vaisala.com/files/An_overview_of_cloud-to-groun-d_lightning_research_in_Brazil_in_the_last_two_decades.pdf).

Reap, R.M., Orville, R.E., 1990. The relationship between network lightninglocations and surface hourly observations of thunderstorms. Mon.Weather Rev. 118, 94–108.

Rivas Soriano, L., De Pablo, F., 2002a. Effect of small urban areas in centralSpain on the enhancement of cloud-to-ground lightning activity. Atmos.Environ. 36, 2809–2816.

Rivas Soriano, L., De Pablo, F., 2002b. Study of lightning event duration andflash rate in the Iberian Peninsula using cloud-to-ground lightning data.Atmos. Res. 61, 189–201.

Rivas Soriano, L., De Pablo, F., Garcia Diez, E., 2002. Relationship betweengeographical latitude and longitude and cloud-to-ground lightning flashcharacteristics in the Iberian Peninsula. Atmosfera 15, 139–146.

Rivas Soriano, L., De Pablo, F., Tomas, C., 2005. Ten-year study of cloud-to-ground lightning activity in the Iberian Peninsula. J. Atmos. Sol. Terr.Phys. 67, 1632–1639.

Schultz, M.D., Underwood, S.J., Radhakrishnan, P., 2005. A method to identifythe optimal areal unit for NLDN cloud-to-ground lightning flash dataanalysis. J. Appl. Meteorol. 44, 739–744.

Schulz, W., Cummins, K., Diendorfer, G., Dorninger, M., 2005. Cloud-to-ground lightning in Austria: A 10-year study using data from a lightninglocation system. J. Geophys. Res. 110 (D09101)10.1029/2004JD005332.

Sonnadara, U., Cooray, V., Götschl, T., 2006. Characteristics of cloud-to-ground lightning flashes over Sweden. Phys. Scr. 74, 541–548.

Tuomi, T.J., 2005. Salamahavainnot 2005 — Lightning Observations inFinland, 2005. Finnish Meteorological Institute. 40 pp.

Tuomi, T.J., Larjavaara, M., 2005. Identification and analysis of flash cells inthunderstorms. Q. J. R. Meteorol. Soc. 131, 1191–121410.1256/qj.04.64.

Tuomi, T.J., Mäkelä, A., 2006. Salamahavainnot 2006 — Lightning Observa-tions in Finland, 2006. Finnish Meteorological Institute. 39 pp.

Tuomi, T.J., Mäkelä, A., 2007. Salamahavainnot 2007 — Lightning Observa-tions in Finland, 2007. Finnish Meteorological Institute. 47 pp.

Tuomi, T.J., Mäkelä, A., 2008a. Thunderstorm climate of Finland 1998–2007.Geophysica 44, 67–80.

Tuomi, T.J., Mäkelä, A., 2008b. Binomial model of lightning detectionefficiency. Journal of Lightning Research 1, 1–8.

Tuomi, T.J., Mäkelä, A., 2008c. Salamahavainnot 2008 — Lightning Observa-tions in Finland, 2008. Finnish Meteorological Institute. 49 pp.

Westcott, N.E., 1995. Summertime cloud-to-ground lightning activity aroundmajor Midwestern urban areas. J. Appl. Meteorol. 34, 1633–1642.