We use ground-based lightning data provided by the lightning locations systems of WWLLN and ENTLN to search the lightning candidates for the forest fires in the periods 2009–2013 and 2014–2019, respectively. In addition, we use lightning optical measurements reported by ISS-LIS between 2017 and 2020 to investigate LCC lightning over Europe.

The ground-based WWLLN is composed by a global network of VLF sensors that can provide the position, time of occurrence, and energy radiation by lightning discharges . WWLLN is more sensitive to CG lightning discharges than IC, with a total global stroke detection efficiency (DE) between 2009 and 2013 of 8 %–13 % . The stroke location accuracy of WWLLN is between 5 and 10 km, while the temporal accuracy is about tens of microseconds . In this work, we use the relationship between the energy of the strokes measured by WWLLN and the peak current of the discharge, as reported by , to estimate the peak current of the fire-igniting lightning.

The ground-based ENTLN is a global network composed of VLF sensors that provide the position, time of occurrence, polarity, and peak current of lightning strokes. ENTLN has a DE of about 90 % for CG strokes over the USA and a total global stroke DE of about 57 % . The median stroke location error is 631 m . In this work, we use the flash product provided by ENTLN. This product is based on the flash criteria proposed by to cluster these strokes into flashes, where two strokes are part of the same flash if they occur in a 0.7 s temporal window and in a 10 km spatial window.

After the end of operation of LIS on the TRMM satellite in 2015, a similar instrument was placed on the ISS for a 2- to 4-year mission, starting in March 2017, covering latitudes between 54.3∘ N and 54.3∘ S . LIS detects optical emissions from lightning with a frame integration time of 1.79 ms and with a spatial resolution of 4 km . LIS assorts contiguous events into groups and clusters groups into flashes with a temporal criteria of 330 ms and a spatial criteria of 5.5 km . proposed a method to identify LCC lightning flashes from the groups reported by TRMM-LIS. According to , optical emissions detected in five or more consecutive frames (time contiguous groups) that are in the same flash can be classified as a LCC lightning flash. In this work, we use the method proposed by to produce a climatology of LCC lightning flashes over Europe based on ISS-LIS lightning measurements between March 2017 and September 2020. We use 20 ms as the lower limit to classify a flash as a LCC flash in order to avoid introducing CG flashes without continuing currents (typical CG) into our climatology.

The data of fires in Spain is provided by the Spanish Ministerio de Agricultura, Pesca y Alimentación . The database includes fires for the whole of Spain between 1968 and 2015 and for all the regions, except for the region of Castilla y León between 2016 and 2017. The causes of fires are provided when known. LIWs are identified by the Spanish Ministerio de Agricultura, Pesca y Alimentación, using lightning data from the Spanish State Meteorological Agency (AEMET). As the set of forest fires between 2016 and 2017 is not complete, we will exclude them from the analysis, except for calculating the spatial flash density associated with LIWs. The total number of fires in this database between 2009 and 2015 is 92 187, among which 3162 were ignited by lightning (about 3.2 %). In this study, we do not include forest fires over the Canary Islands.

Data of fires in Portugal are provided by the Instituto da Conservação da Natureza e das Florestas . The database includes fires between 1980 and 2015. The cause of most fires is not provided. The total number of fires in this database, between 2009 and 2015, is 163 190. We describe, in Sect. 2.4, a criterion to filter out some LIWs from this data set, thereby obtaining 359 LIWs (about 0.22 %).

Data of fires in the Mediterranean area in France are obtained from the Prométhée database . The database includes fires between 1973 and 2018. The causes of the fires are provided. In this study, we only include fires with reported geographical coordinates. The total number of fires in this database between 2009 and 2015, in which the geographical coordinates are provided, is 797, among which 36 were ignited by lightning (about 4.5 %).

The Hellenic Fire Service collects and publishes fire incidence data from 2000 onwards, after receiving (in the summer of 1998) the responsibility of fire extinction in Greece from the Forest Service, including the number of fires and the corresponding area burned . During the last few years, x and y coordinates of the fire ignition points have also been recorded but unfortunately without recording the fire cause. For 3 years during the period 2017–2019, these data were further checked for providing high positional accuracy and made available to be used here. The total number of fires in this database between 2017 and 2019 is 62 690. We use the criterion described in Sect. 2.4 to filter out some LIWs from this data set, thus obtaining 1999 LIWs (about 3.2 %).

We analyze the meteorological conditions of thunderstorms favoring fires or LCC lightning using meteorological data from the European Centre for Medium-Range Weather Forecasts (ECMWF) fifth generation reanalysis ERA5 and the Cloud Top Height product (CTHP) provided by EUMETSAT . In addition, for some selected cases, we analyze the meteorological products provided by the Spanish State Meteorological Agency AEMET;.

ERA5 provides 1 h meteorological data using a 4D-Var assimilation scheme at 139 pressure levels with a horizontal resolution of 0.25∘. The product ERA5-Land provides meteorological data over land by replaying the land component of the ECMWF ERA5 climate reanalysis with an horizontal resolution of 0.1∘ . In this work, we analyzed the 1 h ERA5 meteorological data for cells and time steps containing lightning flashes reported by WWLLN and ENTLN in the Iberian Peninsula and Mediterranean France, for the period 2009–2015, and in Greece, for the period 2017–2019. In particular, we have included in the analysis ERA5 meteorological data for 133 147 and 78 701 cells per hour over, respectively, the Iberian Peninsula and Greece. In addition, we analyzed the ERA5 thermodynamical, dynamical, and microphysical variables for 8351 cells per hour, including lightning flashes reported by ISS-LIS over Europe between 2017–2019. The selected meteorological variables are the convective available potential energy (CAPE), the cloud base height (CBH), the horizontal wind components, the hourly accumulated precipitation, the specific humidity at 450 hPa and vertical profiles between ground and 200 hPa pressure level of the temperature, the relative humidity, the vertical velocity, the specific cloud ice water content, the specific cloud liquid water content, the specific rain water content, the specific snow water content, and the runoff index. The runoff index is the depth (in meters) the water would have if it were spread evenly over the grid box of ERA5 reanalysis, thus representing the amount of water contained in the soil.

AEMET provides meteorological measurements upon request. In this work, we selected some fire-igniting lightning flashes and some LCC lightning over Spain and collected radar echo top height measurements provided by AEMET (maximum height of the 12 dBZ echo). These heights correspond to the maximum heights populated by the precipitable particles.

We use the land cover map from 2015, produced by the European Space Agency (ESA) Climate Change Initiative (CCI) from the PROBA-Vegetation (PROBA-V) and Sentinel-3 OLCI (S3 OLCI) time series, to classify lightning flashes over coniferous and mixed forests. This classification of lightning flashes ensures that only lightning flashes occurring over areas with vegetation that can be ignited are included in our analysis, as most forest fires are due to the ignition of conifers or shrublands during the summer season . This map has a temporal resolution of 1 year and a horizontal resolution of 300 m. We consider that a lightning flash is taking place over coniferous or mixed forests if there is at least one 300m×300 m grid cell with conifer or mixed trees 10 km around the flash position. In particular, we consider coniferous and mixed forests that include the label “needleleaved” in the D3.3.12-v1.3_PUGS_ICDR_LC_v2.1.x_PRODUCTS_v1.3 product. We use elevation data from the NASA Shuttle Radar Topographic Mission to obtain the elevation of LIWs.

The CTHP provided by EUMETSAT is based on measurements of the Meteosat Second Generation (MSG) satellites. The geostationary orbit of MSG satellites is centered at 0∘ N, 0∘ E so that data are reported at the rate of one Earth full-disk scan every 15 min. We investigate the CTH of fire-igniting thunderstorms because it is closely related to the level of convection in thunderstorms and the occurrence of lightning . The CTHP is calculated by EUMETSAT from data acquired by the Spinning Enhanced Visible and InfraRed Imager (SEVIRI) instrument on board the MSG satellites, with a horizontal resolution of 4 km at the center of the orbit and a vertical resolution of 320 m that reaches a maximum altitude of 16 km. The CTHP has been provided every 15 min since June 2011, with a lower rate between 2009 and May 2011 . In this work, we collect all the possible CTH values for the lightning flashes reported by WWLLN and ENTLN in the Iberian Peninsula (and Mediterranean France) between 2012 and 2015. We also collect the CTH values for all the lightning flashes reported by ENTLN in Greece for the period 2017–2019.

We search the most probable CG lightning candidate for each fire using the proximity index A, proposed by , as follows: 1A=1-TTmax×1-DDmax.

The proximity index A combines the distance between each fire and each lightning discharge (D) with the delay between them (T), which is also known as holdover. Parameters (Tmax) and (Dmax) correspond to the maximum holdover and distance between a fire and a lightning discharge to consider the latter as the cause of ignition. We set Tmax=14 d and Dmax=10 km . In the case of no CG lightning discharges taking place within the proposed spatiotemporal window with respect to a fire, we consider that the fire was not ignited by lightning. However, the proximity index can be greater than 0 for fires that were not ignited by lightning.

In the case of the database of fires taking place in Spain and Mediterranean France, the cause of ignition is provided. Therefore, we can easily discard the fires with A>0 that were not ignited by lightning. However, the databases of fires in Portugal and Greece do not provide the cause of ignition, and the proximity index can be greater than zero if a lightning discharge preceded a fire, even if the discharge did not cause the ignition. In order to discard the fires of Portugal and Greece that were not ignited by lightning, we estimate that the threshold value of the A index ensures that at least 80 % of the selected fires are ignited by lightning. To this end, we use the Spanish and French databases of fires. We plot, in Fig. 1, the percentage of fires in Spain and France ignited by lightning using different values of A as a threshold. According to this figure, setting the minimum value of A as 0.7 ensures that 80 % of the selected fires are produced by lightning. Therefore, in our analysis of LIWs over Portugal and Greece, we only consider fires with A≥0.7. This approach can exclude a significant number of LIWs from the analysis. However, it ensures that the sample is not influenced by fires that were not ignited by lightning. Following this approach, we find 359 (0.22 % of the total) and 1999 (3.2 % of the total) LIWs with A≥0.7 in Portugal and Greece, respectively. We have obtained a lower percentage of LIWs in Portugal, with respect to all fires, than in Greece, Spain, and Mediterranean France due to the low DE of WWLLN. Another interesting feature of Fig. 1 is that the percentage of LIWs in the sample with A values lower than 0.3 is higher in the period of 2014–2015 than in 2009–2013. This difference can be explained by the higher DE of ENTLN than WWLLN in the Iberian Peninsula.

As explained in Sect. 2.3, we will limit our analysis to LIWs occurring between May and September and in coniferous and mixed forest.

The standard ISS LIS product includes lightning data at event, group, and flash level . Groups are clustered into flashes if they occur within 330 ms and 5.5 km . In this work, we use group and flash data. We apply the method proposed by for TRMM LIS to classify lightning flashes reported by ISS-LIS as LCC lightning flashes. We consider a flash as a LCC lightning flash if its optical emissions are detected in 10 or more consecutive frames, which means that the detected optical signal is continuous for about 20 ms. In addition, we have classified flashes with optical emissions detected in five or more consecutive frames (optical signal is continuous during about 10 ms; ) to investigate the possible relationships between the duration of the current and the meteorological conditions.

We compare the meteorological variables of fire-igniting lightning flashes to typical CG lightning flashes over coniferous and mixed forest between May and September to identify the characteristics of fire-igniting flashes. In addition, we compare the meteorological conditions associated with LCC lightning flashes reported by ISS-LIS over land in Europe between May and September 2017–2020 to the meteorological conditions of typical lightning flashes to search for possible relationships between meteorology and the occurrence of LCC lightning flashes. The comparison of the meteorological variables is performed as follows:

We collect the 1 h values of the meteorological variables of every ERA5 grid cell containing lightning flashes. For forecasting purposes, we collect the maximum value 3 h before the discharge, as the value of CAPE during the passage of a thunderstorm is significantly lower than the value of CAPE during the development of the storm. For every lightning discharge, we collect the CTH value provided by EUMETSAT.

The number of LIWs over coniferous and mixed forest included in our analysis over Spain, Portugal, France, and Greece are 3162, 359, 36, and 1999 (Table 1), respectively. In Fig. 2, we plot two maps showing the Iberian Peninsula and Greece, the position of coniferous/mixed forests, and the position of all the LIWs included in this study.

From this point, we refer to the contiguous territory, including Spain, Portugal, Mediterranean France, and the Balearic islands, as the Iberian Peninsula. Figures 3a and 4a show the monthly total number of LIWs between May and September in the Iberian Peninsula and in Greece, respectively. In the Iberian Peninsula, the occurrence of LIWs reaches its peak in July, while in Greece the peak occurs in June and a secondary peak occurs in September.

Figures 3b and 4b show the frequency distribution of the lightning peak current of all the lightning taking place over coniferous and mixed forests and fire-igniting lightning during May and September for, respectively, the Iberian Peninsula and Greece. We do not include peak currents during 2009 and 2013, as the WWLLN lightning data do not include information about the polarity and the peak current of the flash. According to these figures, there is a slight shift to higher peaks with a positive polarity. The lack of positive lightning flashes reported by ENTLN with peak currents below 20 kA are a consequence of the data set that do not include positive CG flashes with peak currents below nearly 20 kA. Positive flashes with peak currents below nearly 20 kA are classified as IC .

Figures 3c–d and 4c–d show the distance and holdover between when the reported fire ignition and the lightning candidates for the Iberian Peninsula and Greece, respectively. In the Iberian Peninsula, the 25th, 50th, and 75th percentiles of the distance are respectively 1.30, 2.71, and 5.39 km, while in Greece they are 0.56, 0.93, and 1.48 km. The distance between fires and fire-igniting flashes is significantly lower in the case of Greece, possibly due to improvement in the detection efficiency of ENTLN between the periods 2014–2015 and 2017–2019. The 25th, 50th, and 75th percentiles of the holdovers in the Iberian Peninsula are, respectively, 0.48, 8.13, and 29.23 h, while they are 14.79, 29.43, and 48.78 h in Greece. Differences in the holdover between the Iberian Peninsula and Greece can be due to differences in the detection of fires by local authorities. Figures 3d and 4d suggest a diurnal cycle in the holdover that can be due to different conditions for arrival during the day, similar to what was obtained by for LIWs over Catalonia. At noon, meteorological conditions can favor a rapid arrival after ignition (high temperature and low humidity). Ignitions separated from noon occur under meteorological conditions that do not favor rapid arrival.

We have found that the mean elevation of LIWs in the Iberian Peninsula is 775 m, while in the case of Greece the mean elevation is 319 m.

We investigate the meteorological conditions of LIWs using 1 h ERA5 reanalysis data. We compare several meteorological variables of fire-igniting lightning with typical CG lightning taking place over coniferous and mixed forest between May and September to find the meteorological conditions that favor the ignition of fires by lightning.

Figures 5a and 6a show the frequency distributions of CAPE for the CG lightning and the fire-igniting lightning climatologies for the Iberian Peninsula and Greece, respectively. We have selected the maximum value of CAPE 3 h before the occurrence of lightning. Comparison of both distributions indicates that high values of CAPE are more frequent in fire-igniting flashes than in typical CG lightning flashes. However, according to , lower tropospheric thunderstorm indices, such as CAPE, are not enough to forecast the potential of dry thunderstorms. Nevertheless, the p values lower than 0.05 suggest that local differences in the distributions of CAPE plotted in Figs. 5a and 6a are statistically significant.

Figures 5b and 6b show the frequency distribution of the hourly accumulated precipitation for the CG lightning and the fire-igniting lightning climatologies. In accordance with previous studies over the USA e.g.,, our results also suggest that thunderstorms with a low precipitation rate are associated with a higher probability of lightning-ignited fire production over Europe. In the Iberian Peninsula, the 25th, 50th, and 75th percentiles of the hourly accumulated precipitation for LIWs are, respectively, 0.10, 0.63, and 2.01 mm, while for Greece the percentiles are, respectively, 0.13, 0.88, and 2.76 mm. Again, the p values suggest that differences in the distributions of the accumulated precipitation plotted in Figs. 5b and 6b are statistically significant.

defined lightning superbolts as WWLLN strokes with energy above 106 J in the VLF band between 5 and 18 kHz. We have analyzed the stroke energies in the VLF band between 5 and 18 kHz for WWLLN data in the Iberian Peninsula. We have found 95 superbolts over coniferous and mixed forest and between May and September (among 207 116 strokes) and three superbolts that are also fire-igniting lightning candidates (among 1697 fire-igniting lightning candidates). Therefore, 0.18 % of fire-igniting lightning candidates are superbolts, while 0.05 % of all strokes in coniferous and mixed forest are superbolts. We can then conclude that the incidence of superbolts is higher for detected fire-igniting lightning than for typical lightning, although superbolts do not represent a significant portion of fire-igniting lightning.

We plot, in Figs. 5c and 6c, the comparison of the frequency distribution of the horizontal wind at the surface for the CG lightning and the fire-igniting lightning climatologies. Differences in the median values suggest that stronger winds favor the arrival of fires (distribution shifted to the right) in the Iberian Peninsula and Greece. The p values suggest that differences in the distributions of the horizontal winds plotted in Figs. 5c and 6c are statistically significant.

Figures 5d–f and 6d–f show the frequency distribution of, respectively, the relative humidity (RH) at 850 hPa level, the air temperature at 2 m altitude, and the air temperature at 850 hPa level pressure for the CG lightning and the fire-igniting lightning climatologies. The lower RH and the higher temperature at 850 hPa and 2 m altitude are characteristic conditions of dry thunderstorms over the USA and Australia . These conditions favor the evaporation of precipitable water before it reaches the ground , increasing the probability of fire ignition, survival, and arrival in the Iberian Peninsula and Greece. Differences in the distributions of RH and temperatures plotted in Figs. 5d–f and 6d–f are statistically significant, as suggested by p values lower than 0.05. However, Fig. 6d and f suggest that the RH at 850 hPa level and the air temperature at 850 hPa level pressure would not be good predictors for LIWs in Greece, as the plotted distributions of fire-igniting lightning and CG lightning are not substantially different. Differences in RH and temperature at 850 hPa between Greece and the Iberian Peninsula could be due to topographic effects. LIWs in the Iberian Peninsula tend to occur at higher elevation than in Greece (775 and 319 m, respectively). Therefore, meteorological conditions at 850 hPa would have a greater influence on the ignition, survival, and arrival of fires in the Iberian Peninsula than in Greece.

We have also calculated the median value of the variables plotted in Fig. 6, using A≥0.8 and A≥0.85. We have obtained the same results as when using A≥0.7 in terms of difference between meteorological variables of fire-igniting lightning and typical lightning. However, the p values are significantly higher, as the total number of fire-igniting lightning events included in the sample decreases significantly when increasing the minimum value of A.

The temperature difference between vertical levels can be used to investigate the instability of the atmosphere. We have found that the median temperature differences between 2 m altitude and 700 hPa pressure level for CG lightning in the Iberian Peninsula and Greece are, respectively, 16.8 and 19.0 K. In the case of fire-igniting lightning, the median temperature differences between 2 m altitude and 700 hPa pressure level are 18.8 and 19.9 K, respectively. These results indicate that the instability of the lower troposphere is higher in fire-igniting lightning flashes than in typical CG lightning flashes, as suggested by the analysis of CAPE (see Figs. 5a and 6a).

Figure 7 shows the frequency distribution of the CBH for lightning flashes over coniferous and mixed forest and for all the fire-igniting flashes, indicating that clouds with a high base favor the inception of LIWs. Clouds with a high base favor the evaporation of precipitation before it reaches the ground surface, increasing the probability of survival and arrival of LIWs. Clouds with a high base and low content of moisture in the low or mid level are the typical meteorological conditions of dry thunderstorms in the western USA, as demonstrated by and . Our results suggest, for the first time, that clouds with a high base favor the occurrence and arrival of LIWs in the Iberian Peninsula and Greece. The p values suggest that differences in the distributions of CBH plotted in Fig. 7 are statistically significant.

Finally, we have calculated the median value of the runoff index for fire-igniting lightning and for CG lightning flashes over coniferous and mixed forest. For the Iberian Peninsula, the median value of the runoff index for typical CG flashes is 9×10-5 m, while it is 2×10-5 m for fire-igniting flashes. For Greece, the median values of the runoff index for fire-igniting lightning and for CG lightning are 2×10-4 and 8×10-5 m, respectively. Therefore, the amount of water contained in the soil and, consequently, the state of vegetation can also influence the occurrence of LIWs.

Lightning flash frequency and density are used as proxy variables to forecast or parameterize the inception of lightning-ignited fires. For example, parameterized the inception of LIWs in a global model using the flash frequency, while found that the majority of flash densities were less than 0.41 flashes per square kilometer in lightning-ignited fire days in the USA. In this section, we analyze the temporal evolution and the spatial characteristics of fire-producing thunderstorms in Spain between 2014 and 2015. We assume that the fire-producing thunderstorm is the one that contains the lightning candidate. However, identifying the thunderstorm containing a flash reported by a lightning network is not a trivial problem. proposed the use of the ST-DBSCAN (density-based spatial clustering of applications with noise) algorithm to cluster lightning flashes reported by WWLLN into thunderstorms. The ST-DBSCAN algorithm defines a cluster of flashes as the group of flashes that are neighbors according to selected spatial and temporal thresholds. reported that ST-DBSCAN produces a reliable daily global number of thunderstorms (660±70 on any given moment) by using 0.12∘ in latitude and longitude as a spatial parameter and 18 min as a temporal threshold to cluster WWLLN flashes into thunderstorms. In this work, we cluster lightning flashes reported by ENTLN in the region of interest by searching adequate clustering thresholds by using radar measurements of thunderstorms.

Radar echo top measurements at 12 dB provide the maximum height of precipitable particles . Areas where the height of precipitable particles is high are related to areas with strong updrafts that can favor the production of thunderclouds. We use radar echo top maps to identify the position of thunderstorms before clustering flashes into thunderstorms. Figure 13a and c show two echo top radar images around two fire-ignited lightning events taking place on the Mediterranean coast of the Iberian Peninsula on 17 June 2015 at 18:00 UTC and on 15 June 2015 at 14:15 UTC, respectively. Figure 13b and d show all the lightning flashes reported by ENTLN on the eastern coast of the Iberian Peninsula during 30 min on 17 June 2014 at 18:00 UTC and on 15 June 2015 at 14:15 UTC. Different colors indicate the clusters after applying the clustering algorithm ST-DBSCAN with the spatial and temporal parameters of 0.12∘ and 15 min, respectively. A comparison of both columns in Fig. 13 shows that the clustering parameters of 0.12∘ and 15 min can be used to identify thunderstorms from lightning measurements in the region of interest. used 18 min for WWLLN global lightning data. However, we found that using 18 min for ENTLN lightning data over the Iberian Peninsula would cluster all the lightning flashes shown in Fig. 13b and d into the same thunderstorm.

Following this approach of clustering lightning flashes into thunderstorms, we have clustered the lightning flashes taking place in the Iberian Peninsula between 2014 and 2015 during some days with LIWs. Given that the clustering process requires high computational resources, we restrict our analysis to the period between 2014 and 2015. In order to select the cases of our study, we have only included fire-igniting thunderstorms containing a lightning candidate with a proximity index (A) greater than 0.95 with respect to the fire. Following these criteria, we have selected 77 cases. Figure 14 shows the cumulative number of flashes occurring in the selected cases 1400 min around the ignition of LIWs. Ignitions tend to occur at the moment when the lightning activity just starts to increase from the regime with the lowest lightning activity to the regimen with the highest lightning activity. The regime with a higher lightning activity coincides with a higher precipitation rate that contributes to inhibiting the possibility of fire ignition. We have found no relationships between the cumulative number of the different types and polarities of lightning discharges and the time of ignition.

Several authors have investigated the spatial density of CG flashes 24 h prior to the ignition as a proxy for the occurrence of fires in the USA, depending on the 100 h fuel moisture information e.g.,. They have found a that the threshold value of the CG spatial flash density to ignite a fire and allow it to reach the arrival phase would strongly depend on the lightning detection system. In this work, we have calculated the CG spatial flash density for 5 km around 24 h before the ignition of all the LIWs in Spain during 2016 and 2017, using CG flashes reported by ENTLN. The DE of ENTLN could have increased during recent years. We have calculated the CG spatial flash density for the period between 2016 and 2017 for the case of utility, as the most recent years in our fire database in the Iberian Peninsula are 2016–2017. Therefore, is the most recent information. We obtain a median value of 0.44 flashes per square kilometer, a 25th percentile of 0.26 flashes per square kilometer, and a 75th percentile of 0.69 flashes per square kilometer. In addition to meteorological parameters, these values could be used in regional fire forecasting or atmospheric models to quantify the risk of ignition by thunderstorms, especially if radar measurements of the precipitation rate are not available.

In this section, we investigate the climatology of LCC lightning flashes over Europe derived from ISS-LIS data between March 2017 and September 2020 and the meteorological and cloud conditions under which they take place. This approach will help us to identify possible relationships between LCC lightning and fire-igniting lightning flashes.

Let us now compare the meteorological and cloud conditions of fire-igniting lightning and LCC(>20 ms) lightning. According to Fig. 18, there exists a subset of LCC(>20 ms) lightning flashes taking place in clouds with a high base (CBH larger than 2000 m). As we discussed in Sect. 3.2, high-based thunderclouds favor the occurrence of dry lightning and LIWs. In particular, Fig. 8 shows that there is a significant total number of fire-igniting flashes taking place in thunderstorms with CBH larger than 2000 m. For this reason, in this section we analyze the meteorological conditions of LCC(>20 ms) lightning flashes taking place in clouds with a high base with CBH values larger than 2000 m. The total number of LCC(>20 ms) lightning flashes in high-based thunderclouds is 392. LCC(>20 ms) lightning flashes with CBH>2000 m are plotted as red dots in Fig. 15a.

Figure 20 shows the frequency distribution of some meteorological variables for typical lightning and LCC(>20 ms) lightning flashes in high-based thunderclouds. According to this figure, the median values of accumulated precipitation and RH at 850 hPa are lower for LCC(>20 ms) lightning flashes in high-based thunderclouds than for the climatological median, with p values lower than the threshold of 0.05. In addition, temperatures at 2 m altitude and at 850 hPa are higher for LCC(>20 ms) lightning flashes in high-based thunderclouds than for the climatological median, again with p values lower than the threshold of 0.05, indicating that differences are statistically significant. These are typical conditions of fire-igniting lightning flashes (see Sect. 3.2). We have found slightly lower values of CAPE for LCC(>20 ms) lightning flashes in high-based thunderclouds than for typical lightning.

A comparison of Figs. 10 and 17 shows that fire-igniting lightning flashes in the Iberian Peninsula and LCC(>20 ms) lightning flashes take place in thunderstorms with weaker updrafts than the climatological median. We do not find the same tendency in Greece (see Fig. 10). As we show in Figs. 8 and S6 in the Supplement, the content of the moisture of LIWs-producing thunderstorms in the Iberian Peninsula and LCC(>20 ms) lightning-producing thunderstorms is lower than the climatological median, which could be the reason behind the weaker updrafts, as drier air is more dense than moist air . Deep convective clouds are usually associated with the 440 hPa level , which is included in the region where the updraft is weaker than the climatological median for LIWs in the Iberian Peninsula and LCC(>20 ms) lightning. Several lightning parameterizations are based on the updraft mass flux at 440 hPa, where electrification occurs . Weaker updraft leads to a slower rime accretion rate . As suggested by , a weaker charging rate can lead to the development of larger charge regions before lightning is initiated, producing a larger charge region available to be neutralized by a lightning flash. Large charge regions can favor the occurrence of LCC(>20 ms) lightning and, according to our results, could also favor the occurrence of LIWs.

We have compared the LCC(>20 ms) lightning flashes with CBH>2 km taking place in the Iberian Peninsula and Greece with ENTLN flash data. Among 39 LCC(>20 ms) lightning flashes reported by ISS-LIS, we have found 12 ENTLN lightning flashes (3 CG and 9 IC) taking place in a radius of 0.2∘ (latitude and longitude) around the flashes reported by ISS-LIS and within a temporal delay smaller than 2 s. Therefore, we conclude that ENTLN is not adequate to detect the continuing phase of LCC(>20 ms) flashes in the studied regions. These results are in agreement with , who reported that between 25 % and 40 % of LCC lightning flashes detected by TRMM-LIS only have an IC pulse detected by the National Lightning Detection Network (NLDN), with no CG strokes detected.

We also attempted to directly correlate LIWs in Greece (2017–2019) and over the Iberian Peninsula (2017) with LCC lightning flashes reported by ISS-LIS. However, we did not find any case in which ISS-LIS was flying over Greece or the Iberian Peninsula at the moment of ignition.