Tools for validation of the Lightning Imager sensor on the 3rd generation of the METEOSAT weather satellite

Master’s Degree in Space and Aeronautical Engineering Tools for validation of the Lightning Imager sensor on the 3rd generation of the METEOSAT weather satellite Master Thesis Author: M. Núria Partal Camps Supervisor: Joan Montanyà Puig Spring term 2019/20 Index Declaration of honour ...................................................................................... 6 Abstract ........................................................................................................... 7 1. Lightning Detection Systems .......................................................................... 8 1.1. Lightning Mapping Array .......................................................................... 9 1.1.1. System characteristics and operation ................................................ 9 1.1.2. Colombia Mapping Array COL-LMA ................................................. 10 1.2. GOES-R Geostationary Lightning Mapper ............................................. 12 1.2.1. GOES-R Satellites ........................................................................... 12 1.2.2. System characteristics and operation .............................................. 13 1.2.3. Events, groups and flashes .............................................................. 15 1.3. Meteosat MTG Lightning Imager ............................................................ 16 1.3.1. MTG Satellites ................................................................................. 16 1.3.2. System characteristics and operation .............................................. 16 2. Methodology ................................................................................................. 18 2.1. Data files ................................................................................................ 18 2.1.1. Lightning Mapping Array Data ......................................................... 18 2.1.2. Geostationary Lightning Mapper Data ............................................. 19 2.2. Data Analysis ......................................................................................... 20 2.2.1. Pre-processing: LMA and GLM data normalization.......................... 20 2.2.2. Comparison parameters .................................................................. 20 3. Results ......................................................................................................... 22 3.1. Cases evaluated .................................................................................... 22 3.2. Flash Detection Efficiency with LMA as reference ................................. 23 3.2.1. Detection Efficiency Parameter Results ........................................... 24 3.2.2. Qualitative maps .............................................................................. 27 3.2.3. Flash False Alarms .......................................................................... 29 3.2.4. Distance between correlated flashes centroids ................................ 31 3.2.5. Number of GLM events and flashes per LMA flash .......................... 33 3.2.6. Distribution of number, power and height of LMA sources for flashes detected and not detected by GLM ............................................................ 34 3.2.7. Detection Efficiency vs maximum height of LMA flashes ................. 36 3.3. Flash Detection Efficiency with GLM as reference ................................. 36 3.4. Flash Duration ........................................................................................ 41 3.5. Location accuracy .................................................................................. 41 4. Conclusions .................................................................................................. 44 Bibliography ...................................................................................................... 46 Table of figures Figure 1. Geographical location of Colombia Lightning Mapping Array network COL-LMA in Colombia. Upper image is for old location in Santa Marta [7], lower image is for the new location in Barrancabermeja. .......................................... 11 Figure 2. Outline of the GOES-16 and GOES-17 position and region coverage [10]. .................................................................................................................. 13 Figure 3. Right, sketch of the sensor unit and electronics unit of GLM [11]. Left, GLM being prepared to undergo vibration testing [12]. .................................... 14 Figure 4. Illustration of a GLM flash formed by 2 groups and 20 events. The dots represent LMA sources [11]. ..................................................................... 15 Figure 5. Field of View of the Lightning Imager instrument. [14] ...................... 17 Figure 6. NetCDF data file structure for a GLM file. ......................................... 19 Figure 7. Preview of LMA data to evaluate in scenario A (left) and B, C and D (right) ................................................................................................................ 22 Figure 8. Qualitative maps for LMA sources and GLM events for case A. ..... 27 Figure 9. Qualitative maps for LMA sources and GLM events for case B. ..... 27 Figure 10. Qualitative maps for LMA sources and GLM events for case C. ... 28 Figure 11. Qualitative maps for LMA sources and GLM events for case D. ... 28 Figure 12. Qualitative maps for GLM events not correlated to a LMA flash. ... 30 Figure 13. Histogram and boxplot for the distance values of all the flashes correlated. ........................................................................................................ 31 Figure 14. Two examples of LMA flashes sharing one GLM flash. The ones in the left are both separated from the GLM flash by less than 25 km, while one of the ones in the right is separated by 27 km. ..................................................... 32 Figure 15. Two examples of LMA flashes sharing one GLM flash. In both examples there is one LMA flash separated from the GLM flash by more than 25 km, but it’s a correct correlation. .................................................................. 32 Figure 16. Two examples of LMA flashes partially sharing one GLM flash. The events relation with each LMA flash can be quickly seen as not all GLM events have the same markers at all times. ................................................................. 33 Figure 17. Right: distribution of the number of GLM flashes per LMA flash. Left: Accumulated number of GLM events per LMA flash. ....................................... 33 Figure 18. Number of LMA sources for flashes detected and not detected by GLM. ................................................................................................................ 34 Figure 19. Power of LMA sources for flashes detected and not detected by GLM. ................................................................................................................ 35 Figure 20. Height of LMA sources for flashes detected and not detected by GLM. ................................................................................................................ 35 Figure 21. Influence of flash height to Detection efficiency. ............................ 36 Figure 22. Right: distribution of the number of LMA flashes per GLM flash. Left: Accumulated number of LMA sources per GLM flash. ..................................... 39 Figure 23. Number of GLM events for flashes detected and not detected by LMA. ................................................................................................................. 39 Figure 24. Radiance of GLM events for flashes detected and not detected by LMA. ................................................................................................................. 40 Figure 25. Influence of flash radiance (energy) to Detection efficiency. ......... 40 Figure 26. Density of data in grid in percentage units for case A. Left: LMA sources density. Right: GLM events density..................................................... 42 Figure 27. Density of data in grid in percentage units for case B. Left: LMA sources density. Right: GLM events density..................................................... 42 Figure 28. Density of data in grid in percentage units for case C. Left: LMA sources density. Right: GLM events density..................................................... 43 Figure 29. Density of data in grid in percentage units for case D. Left: LMA sources density. Right: GLM events density..................................................... 43 Table of tables Table 1. GLM performance characteristics [11]. .............................................. 14 Table 2. Data of cases evaluated. .................................................................... 22 Table 3. Flash detection efficiency parameters for data evaluation ................. 23 Table 4. Detection efficiency for initial parameters ........................................... 24 Table 5. Detection efficiency for Range increase to 150 km ............................ 24 Table 6. Detection efficiency for elimination of minimum sources required to evaluate LMA flash ........................................................................................... 25 Table 7. Detection efficiency for distance between centroids decrease to 25 km ......................................................................................................................... 25 Table 8. Detection efficiency for Range increase to 150 km, elimination of minimum sources required to evaluate LMA flash and distance between centroids decrease to 25 km ............................................................................ 26 Table 9. GLM events and flashes not correlated to a LMA flash. ..................... 29 Table 10. Detection efficiency for established parameters ............................... 37 Table 11. LMA sources and flashes not correlated to a GLM flash .................. 37 Table 12. GLM flashes repeated during LMA flash detection and LMA flashes repeated during GLM flash detection. .............................................................. 38 Table 13. Mean time duration (in seconds) for LMA and GLM flashes and those flashes associated to a flash for analysis in sections 3.2 and 3.3. .................... 41 6 Declaration of honour I declare that, the work in this Master Thesis is completely my own work, no part of this Master Thesis is taken from other people’s work without giving them credit, all references have been clearly cited. I’m authorised to make use of the research group related information I’m providing in this document. I understand that an infringement of this declaration leaves me subject to the foreseen disciplinary actions by The Universitat Politècnica de Catalunya - BarcelonaTECH. M. Núria Partal Camps __________________ 09.11.2020 Student Name Signature Date Title of the Thesis: Tools for validation of the Lightning Imager sensor on the 3rd generation of the METEOSAT weather satellite 7 Abstract The 3rd generation of the METEOSAT (MTG) will be equipped with a lightning imager sensor (LI) to detect and locate lightning flashes. The field-of-view of the LI-MTG will cover Europe and Africa. Since the LI-MTG will observe optical emissions from lightning from cloud tops, it will be able to provide total lightning activity in thunderstorms including intra-cloud and cloud-to-ground. At ground, the unique system that detects total lightning with high efficiency is the Lightning Mapping Array (LMA). The UPC Lightning Research Group operates the Ebro LMA network at the Delta de l’Ebre area. This network will be used to validate the MTG LI performance when it will be in operation. To do that, the UPC LRG group has recently compared the ISS-LIS imager sensor on the International Space Station and the needs have been defined. The aim of this project is to follow the comparison between the ISS-LIS imager and LMA to define the needs for LI-MGT. To do that, data from the Geostationary Lightning Mapper (GLM) will be compared with the Colombia LMA at Barrancabermeja, as both LI-MGT and GLM are geostationary satellites. With the results obtained the bases for the final validation tool of MTG LI will be set and also will be reached a better understanding of the data gathered by a lightning detection system on board of a geostationary satellite. 1. Lightning Detection Systems 8 1. Lightning Detection Systems It was Greek philosophers who started what would be the predecessor to what nowadays is meteorology [1]. Aristotle wrote “Meteorological” around 340 B.C, where he tried to deduce some things about weather, and for about 2000 years that was the only work on this subject. In 1593, Galileo invented the thermometer and years later, in 1643, Evangelista Toricelia invented the barometer. From that moment, various inventions were created to measure wind speed, levels of precipitation and other meteorological phenomena. But it wasn’t until the 1740’s that Benjamin Franklin saw similarities between the electricity he was using in his experiments and lightning. This led him to make experiments to prove storm clouds were electrically charged and years later to the invention of the lightning rod [2]. A long time has passed from that days, and now buildings, cars and aircrafts are made taking into account the damage a lightning strike can do, and therefore using the appropriate protective measures that have been created as more information about lightning and their characteristics has been discovered. This information is obtained with different systems that are able to track lightning from the ground or from space and obtain their characteristics. While ground-based lightning locating systems have been used since the 1920s, space-based ones are quite new considering they have been operating for almost two decades [3][4]. Space lightning observation systems have become the key piece to achieve realtime global measurements of lightning activity, making them the future of lightning observation. Nowadays, both ground and space lightning detection systems work together, complementing each other and in numerous occasions ground systems are used as tools for validation of the new space systems created. In this chapter will be explained three of the systems currently used to obtain information about lightning: the Lightning Mapping Array, the GOES-R Geostationary Lightning Mapper and the Meteosat Lightning Imager Sensor. 1. Lightning Detection Systems 9 1.1. Lightning Mapping Array The Lightning Mapping Array or LMA, is a three-dimensional lightning location system. LMA measures the time of arrival of the very high frequency (VHF) emissions produced by lightning to determine the location of their sources and with that information creates a three-dimension map of the lightning activity. The system was developed by researchers at the New Mexico Institute of Mining and Technology and was originally used in Oklahoma in June 1998. There are various networks currently operating, for example in Europe there are the Ebro LMA (ELMA) and Corsica’s SAETTA; and in USA there are the Oklahoma LMA (OKLMA), the North Alabama LMA (NALMA), the West Texas LMA (WTLMA) among many others. The network studied in this project is located in Barrancabermeja, Colombia (COL-LMA). LMA have a very high detection efficiency and can detect both cloud to ground and intracloud discharges, which makes it a very useful tool to perform calibrations for satellite-based sensors as GLM or the future Meteosat LI. [5] 1.1.1. System characteristics and operation LMA consists in a set of VHF antenna stations (between 7 and 20) surrounding a central station, this station is the one that calculates the location and emission power of the lightning source using the Time of Arrival (TOA) technique. These stations are usually placed over a region of 60-80 km diameter and there is a distance of 15-20 km between each other. The VHF antennas can detect sources at a distance of 200 km around the centre of the antenna network. Lightning discharges radiate over a broad range of radio frequencies, between VLH and VHF; for VHF, they cover from 30 MHz up to 300 MHz. The frequency in which the antennas operate will depend on the usage of VHF frequencies in the region, and the threshold at which they identify a possible source will be determined by the noise expected in the area. As an example, a LMA network located in Ebro’s Delta in Spain operates with frequencies between 60 to 66 MHz, as those belonged to the old TV system and now are unused. Therefore, when in normal operation mode, a VHF antenna is able to detect every 80 µs a source which signal has a magnitude above their established threshold, they send the time when the signal was detected (time of arrival) to the central station. Then the central station calculates the time, altitude, latitude and longitude of the lightning with the already known distances between antennas 1. Lightning Detection Systems 16 1.3. Meteosat MTG Lightning Imager The Ligtning Imager or LI, is an optical instrument that will be on board the Meteosat Third Generation satellite. LI will map the total lightning activity during day and night over more than the 80% of the visible Earth disc. [13] This instrument will complement the Geostationary Lightning Mapper onboard GOES-R and the China Meteorological Administration Lightning Mapper onboard the FY-4. The information provided by the Lightning Imager will be used for nowcasting, aid to air traffic safety and to understand the physical and chemical processes in the atmosphere. 1.3.1. MTG Satellites Meteosat Third Generation or MTG is a cooperative program between EUMETSAT and ESA to produce the next generation of geostationary satellites. This generation will improve their imaging service as it will include a new Lightning Imager and a state-of-the art atmospheric sounding service. The program consists in six geostationary satellites that will launch from 2022 onwards, four of them will be Imaging Satellites (MTG-I) and two will be Sounding Satellites (MTG-S). Their operational configuration implemented will consist of one MTG-S satellite and two MTG-I satellites operating in tandem, one will scan Europe and Africa every 10 minutes and the other one will scan Europe every 2.5 minutes. On board the imaging satellites, MTG-I, will be found the Flexible Combined Imager and the Lightning Imager. On the sounding satellites, MTG-S, an interferometer, the Infrared Sounder and the Copernicus Sentinel-4 will be found. 1.3.2. System characteristics and operation The Lightning Imager will detect energy received from photons. Same as GLM, If the energy detected exceeds the trigger threshold established, it is identified as LI triggered Event. The instrument is formed by four identical cameras, each one covers a different domain of the field of view, as shown in Figure 4. 1. Lightning Detection Systems 17 Figure 5. Field of View of the Lightning Imager instrument. [14] Again, a GLM, the data produced by this instrument is given as a set of events, groups and flashes. Events will be the triggered pixels in the detector grid, groups will be those events in the same integration period that are next to each other and finally flashes will be collections of groups in temporal and spatial vicinity. 2. Methodology 18 2. Methodology In this chapter will be explained the data received from the different lightning detection systems discussed before: Lightning Mapping Array and Geostationary Lightning Mapper. Both systems provide a raw file that is processed before using it in the comparing program made in this project. Both files, raw and processed will be explained and compared to understand why the processing was necessary. The programs created to analyse and compare both systems will also be explained in this chapter, in order to understand the creation process and the capabilities that the program has. 2.1. Data files 2.1.1. Lightning Mapping Array Data As explained in section 1.1.1., each LMA VHF antennas give information of the distance and timing of the sources detected and afterwards the information of each antenna is processed to determine the 3D location of the source. In COL-LMA station, each antenna dumps this information into a “.dat” file every 10 min. With this information, an algorithm made by UPC’s Lightning Research Group crosses all the different antenna data and returns another “.dat” file containing the position, timing, power among other values for each source detected for a 10 min period. With this algorithm it is possible to change the number of antennas to be used for the process. This is important as, a very high number of antennas give less noisy results but it can cause a loss of information, as a lightning event may be detected by less antennas than required. On the other hand, if a low number of antennas is selected, the noise will be higher and more false positive events would be considered as valid. Before using the data, another algorithm also made by the Lightning Research Group finds and identifies the flashes and the noise in the data. This algorithm returns a “.txt” file containing an identifier for noise (0) and lightning flash ID (≠0), time, latitude, longitude, height and power. With this program is also possible to display the events, and has been the reference for the display program made in this project. 2. Methodology 19 2.1.2. Geostationary Lightning Mapper Data NOAA shares the GLM data online on Google Cloud Platform, there can be found the data organized in folders per year, day of the year and hour. Then, files with data for every 200 milliseconds are available to download. This GLM data files have netCDF format “.nc”. NetCDF is a data model capable to store large amounts of array-oriented data in a way that is self-describing and portable. This type of dataset contains dimensions, variables, and attributes, and all have both a name and an ID number in order to be identified. [15] Therefore, the GLM files contain a lot of information that may not be of interest. Figure 5 shows the structure of a GLM netCDF file, it contains different variables, and their descriptions. There is information related to event, flash and group levels but also about the satellite status at that moment (not shown in picture). Figure 6. NetCDF data file structure for a GLM file. In order to extract only the information required for location and identification of lightning, a python code has been developed by UPC’s Lightning Research Group. This program takes all the GLM data files saved in a specified folder and returns a “.txt” file with time (in seconds), group latitude and longitude, flash ID, flash latitude and longitude, and energy for each event. A second processing is needed for this information, as the first results of the data processed showed an error where more than one flash could have the same ID. 2. Methodology 20 In order to avoid this bug, a python algorithm was created to find the problematic IDs, separate the flashes and assign them new unused IDs. 2.2. Data Analysis 2.2.1. Pre-processing: LMA and GLM data normalization In order to compare the data between LMA and GLM, the data of LMA will be taken as reference for time and space delimitation, as GLM covers more space range and if it’s not delimitated some results could be misleading. To delimitate the GLM data in time it will be used the time of the first and last LMA sources. To delimitate the area of evaluation, a fixed value of 150 km from the LMA network centre will be set, which is the double the value of the estimated LMA range used. 2.2.2. Comparison parameters Once GLM data has been delimited it will be evaluated against LMA data. In order to do so, a series of parameters have been defined to evaluate and compare the data correlation between both lightning location systems. This parameters have been determined using as reference a ISS-LIS data analysis based on LMA networks in Europe [16]. The comparison parameters to evaluate are: - Flash Detection Efficiency: A correlation between GLM flashes and LMA flashes based on the distance between their centroids and the time of the events/sources. Once GLM flashes have been assigned to a LMA flash (or not), the Detection Efficiency for GLM (DEf) is calculated following the formula 𝐷𝐸𝑓= 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐿𝑀𝐴 𝑓𝑙𝑎𝑠ℎ𝑒𝑠 𝑑𝑒𝑡𝑒𝑐𝑡𝑒𝑑 𝑏𝑦 𝐺𝐿𝑀 𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐿𝑀𝐴 𝑓𝑙𝑎𝑠ℎ𝑒𝑠 𝑒𝑣𝑎𝑙𝑢𝑎𝑡𝑒𝑑 Using the flash correlations found with these criteria various items are calculated: • Qualitative maps • Flash False Alarm • Number of GLM events and flashes per LMA flash • Distribution of number, power and height of LMA sources for flashes detected and not detected by GLM 2. Methodology 21 • Flash detection efficiency vs maximum height of LMA flashes - Flash duration: An average of the duration of LMA flashes, GLM flashes and of the correlated flashes is made to compare their differences. - Location accuracy: The Detection efficiency flashes are compared using the distance between their centroids, but this could lead to mistakes in the assignation of flashes. To determine the accuracy in the location of correlated flashes, LMA data should be rearranged to fit a 0.075º square grid that would make easier the comparation with the grided data of GLM. With LMA data rearranged in a grid, a series of qualitative maps to visually compare the LMA and GLM are obtained. 3. Results 22 3. Results Following the methodology presented in Chapter 2, in this chapter are explained the results obtained after evaluating the LMA and GLM flash data for four different cases. 3.1. Cases evaluated The LMA and GLM data of four cases are evaluated to determine the correlation existing between the two lightning detection systems. Table 2. Data of cases evaluated. Case Date Time interval (Local time) Sources detected Events detected A 30 Oct. 2019 00:30 - 05:30 697955 318134 B 31 Oct. 2019 20:30 - 23:00 2535 4806 C 01 Nov. 2019 23:40 - 05:40 307376 189356 D 01 Nov. 2019 05:40 - 08:30 104306 42971 Case A is one of the longest in terms of time and is the one with more sources and events detected. This case and case B happen during the night, when it should be easier for GLM to detect events, since the amount of light interfering with the detection of lightning is lower. Cases C and D belong to the same meteorological event, as can be observed in Figure 7. This event is divided in two parts to separate the data detected before and after sunrise, which was at 05:43 local time. It will be interesting to evaluate if there are any differences in the results between C and D, and D, A and B, as the sunlight during those cases will be different and may be affecting the GLM detection. Figure 7. Preview of LMA data to evaluate in scenario A (left) and B, C and D (right) 3. Results 23 3.2. Flash Detection Efficiency with LMA as reference As explained in section 2.2.2., in order to calculate the flash detection efficiency, the GLM flashes are assigned to a LMA flash. In order to do so, various parameters have to be stablished to determine if two flashes are corelated. Five parameters are stablished to discard LMA sources out of range, to define the margins allowed to assign a GLM flash to a LMA flash and to assess the quality of a LMA flash and its detection by GLM. LMA detects sources within a range from the centre of the station, and this range may change depending on the number of stations available or the orography of the surroundings. A good process to set the range would be to create a grid to evaluate the power detection of the LMA data. The cells with lowest power values would indicate the most sensitive areas, which would be the ones were the detection is more precise. Then, with the most sensitive areas detected a good range value can be determined. For this project, an estimated value of 75 km from the centre of the LMA network has been defined as the range of the LMA network. This value ensures that the flashes within this range have been detected in a good efficiency area. To determine if a GLM flash is correlated to an LMA flash it has to be checked if both happen at the same time and at the same location. To compare the location, an initial value for the maximum distance allowed between centroids of both flashes is set to 50 km. In order to allow some margins to the LMA flash time window, a time tolerance is applied to the start and end of the flash. The initial value of this parameter is set to 0.1 seconds. For LMA flashes, it would be interesting to evaluate only flashes with a minimum number of sources per flash, as they would represent larger and better detected flashes. To do so, an initial value for the minimum sources required to evaluate a LMA flash is set to 50. Then, from those LMA flashes that are evaluated, only those who are related to at least one GLM event will be considered detected. A summary of the parameters defined is shown in Table 3. Table 3. Flash detection efficiency parameters for data evaluation Area range for LMA 75 km Max. distance between centroids of GLM and LMA flashes 50 km Time tolerance 1 s Minimum number of LMA sources to accept a flash 50 Minimum number of GLM events to detect a flash 1 3. Results 24 3.2.1. Detection Efficiency Parameter Results With the parameters established in Table 3, the following results are obtained. Table 4. Detection efficiency for initial parameters Case Average flash rate (min-1) Average LMA source rate (s-1) Number of flashes Number of flashes detected by GLM Detection efficiency A 3.85 18.45 2231 2117 0.95 B 0.1 0.11 20 20 1 C 2.14 6.88 1379 1229 0.89 D 0.41 2.1 325 292 0.9 Total number of LMA flashes 3955 Total number of LMA flashes detected 3658 Total Detection efficiency 0.92 For this initial result, the detection efficiencies obtained are really high. In case B, all flashes evaluated are corelated with a GLM flash, probably due to the low number of flashes and the low intensity of the event, as the flash and source rates are very low too. As it was expected, cases C and D, the ones with possible sunlight affecting the GLM detection, have by little the lowest values of detection efficiency. Calculating the detection efficiency for different values of range, allowed distance between centroids and minimum number of sources in flash, the following results are obtained. Table 5. Detection efficiency for Range increase to 150 km Range 150km Average flash rate (min-1) Average LMA source rate (s-1) Number of flashes Number of flashes detected by GLM Detection efficiency A 3.69 17.6 2242 2128 0.95 B 0.1 0.12 22 22 1 C 2.17 6.92 1395 1245 0.89 D 0.41 2.10 325 292 0.9 Total number of LMA flashes 3984 Total number of LMA flashes detected 3687 Total Detection efficiency 0.93 For the first change the range is increased, therefore more LMA sources can be evaluated and more LMA flashes should be detected. This increase in LMA flashes is obtained, but it’s so low that it does not affect the detection efficiency. 3. Results 25 The range has been increased to double the estimated range of the LMA in order to evaluate the change in the detection efficiency but, as explained before, determining a good value for the range is more complicated. Table 6. Detection efficiency for elimination of minimum sources required to evaluate LMA flash No min sources Average flash rate (min-1) Average LMA source rate (s-1) Number of flashes Number of flashes detected by GLM Detection efficiency A 9.26 18.67 5731 4372 0.76 B 0.35 0.17 74 71 0.96 C 5.74 7.86 3701 2885 0.78 D 0.79 2.16 631 505 0.8 Total number of LMA flashes 10137 Total number of LMA flashes detected 7833 Total Detection efficiency 0.77 For the second parameter to change, the requirement of 50 sources to evaluate a LMA flash is removed. This means that all flashes detected, even if they are formed by only one source are evaluated. The number of flashes with this change has increased, in more than two times the initial results in all cases. The flash rate has also increased greatly, while the source rate has done it in a more contained way. But the other side, the detection efficiency has decreased. The number of detected flashes has not increased as much as the number of flashes evaluated, and that could be an indicator that flashes with less than 50 sources could be not detected by GLM or are VHF emissions detected and not related with a flash. In this scenario, the lowest detection efficiency is in A, one of the night cases. Table 7. Detection efficiency for distance between centroids decrease to 25 km Centroid dist. 25 km Average flash rate (min-1) Average LMA source rate (s-1) Number of flashes Number of flashes detected by GLM Detection efficiency A 3.85 18.45 2231 2077 0.93 B 0.1 0.11 20 20 1 C 2.14 6.88 1379 1223 0.89 D 0.41 2.1 325 290 0.89 Total number of LMA flashes 3955 Total number of LMA flashes detected 3610 Total Detection efficiency 0.91 3. Results 32 Figure 14. Two examples of LMA flashes sharing one GLM flash. The ones in the left are both separated from the GLM flash by less than 25 km, while one of the ones in the right is separated by 27 km. For the example in Figure 14, reducing the distance would be beneficial as the flash in the right would be assigned only to the correct flash and will count only once on statistical studies. Figure 15. Two examples of LMA flashes sharing one GLM flash. In both examples there is one LMA flash separated from the GLM flash by more than 25 km, but it’s a correct correlation. But now, in the examples in Figure 15, the LMA flashes do belong to the GLM flash, and if the distance is reduced those correlations would be lost. Another interesting aspect is that some GLM flashes are divided between various LMA flashes, not sharing all the events in both correlations but only part of them. 3. Results 33 Figure 16. Two examples of LMA flashes partially sharing one GLM flash. The events relation with each LMA flash can be quickly seen as not all GLM events have the same markers at all times. Since GLM not only merges events in flashes but also groups, it would be interesting to make an assignation of GLM groups to LMA flashes to see if those have a higher detection efficiency decreasing the distance between centroids. 3.2.5. Number of GLM events and flashes per LMA flash Contrary to the results for ISS-LIS, for GLM it is not usual to have more than one GLM flash corelated with a LMA flash. Figure 17. Right: distribution of the number of GLM flashes per LMA flash. Left: Accumulated number of GLM events per LMA flash. 3. Results 34 This also makes sense with the found GLM flashes related to various LMA flash, as it indicates that it may be possible that one GLM flash is formed by various LMA flashes. This change between ISS-LIS detection and GLM detection could be due to the difference between being in a low orbit, were the field of view is changing and possibly the identified flashes need to be shorter, and being in a geostationary orbit, that would not limit the time window for detection of flashes. 3.2.6. Distribution of number, power and height of LMA sources for flashes detected and not detected by GLM In order to see it there are any significant differences between those LMA flashes detected and not detected, some statistics for the number of sources, power and height of the flashes has been computed. Figure 18. Number of LMA sources for flashes detected and not detected by GLM. For the number of sources, it seems like those flashes that are detected have a higher number of sources. The median for detected is 162.5 and for not detected is 94. 3. Results 35 Figure 19. Power of LMA sources for flashes detected and not detected by GLM. In terms of VHF power, the values obtained for detected and not detected flashes are quite similar. The median is 15.9 and 13.3 respectively. Figure 20. Height of LMA sources for flashes detected and not detected by GLM. Height has the biggest difference between detected and not detected mean, being 9591 and 7101 respectively. This difference could indicate that deep clouds on top of a lightning could largely affect the detection of this flash from space. In order to have another look to the possible effect of flash height in GLM flash detection, the detection efficiency is calculated for various height ranges in the next section. 3. Results 36 3.2.7. Detection Efficiency vs maximum height of LMA flashes In order to see the influence of flash height in flash detection, the Detection efficiency has been calculated based on their height, using the maximum height value as reference. Figure 21. Influence of flash height to Detection efficiency. The heights in the cases studied are quite high, as the average value is 0.7 times the maximum height, this results in a slow decrease of the detection efficiency for heights above 0.5. For heights under 0.25, there are no flashes to evaluate. 3.3. Flash Detection Efficiency with GLM as reference After seeing the results of sections 3.2.3 and 3.2.5, it seems as GLM flashes could be assigned to more than one flash. For this reason, it seems interesting to perform the flash detection efficiency procedure again using GLM flashes as reference. Therefore, in the next results, the assignation of LMA flashes to GLM flashes has been done with the same parameters as established before. GLM flashes will only be evaluated if they contain more than 50 events, distance between centroids of the correlated flashes will be less than 50 km, and with one source the GLM flash will be considered detected. 3. Results 37 Note that all the GLM data used is inside the 75 km range of the LMA centre, as no region delimitation can be applied to LMA with GLM restrictions (as GLM covers more area than LMA), and that way the GLM is more restricted and the detection efficiency better represented. Table 10. Detection efficiency for established parameters Case Average flash rate (min-1) Average GLM event rate (s-1) Number of flashes Number of flashes detected by LMA Detection efficiency A 6.54 14.42 1954 1386 0.71 B 0.46 0.84 31 25 0.81 C 3.71 9.17 1126 1058 0.94 D 1.37 4.28 208 207 0.99 Total number of GLM flashes 3319 Total number of GLM flashes detected 2676 Total Detection efficiency 0.80 Again, the best results of detection efficiency are obtained for the initial values (Table 3), and the next results are obtained using them. Qualitative maps are omitted, as they do not provide any additional information than that commented in section 3.2.2. For LMA flashes not correlated to GLM flashes, the following results are obtained. Table 11. LMA sources and flashes not correlated to a GLM flash Case LMA sources Not correlated (%) LMA flashes Flashes not correlated (%) A 712435 196249 28 5974 3649 61 B 2535 1144 45 96 69 72 C 311581 93001 30 3922 2006 51 D 106242 38037 36 638 310 49 Total number of LMA flashes 10630 Total number of not correlated flashes 6034 Percentage 57 As with LMA, an important part of the flashes are not correlated. In this case, more than the 50% of the flashes are not correlated, while events not reach the 50%. The quantity of LMA data outside the range made for GLM is less than the GLM data in the previous case, therefore less flashes are out of range and with no possibility for a range. Repetition of LMA flashes on various GLM flashes is also found. Discarding the ones were the problem could be solved with a reduction of the maximum distance 3. Results 38 between centroids, this time the interesting aspect of these repeated flashes is that the GLM flashes that share a LMA flash are coincident in time. During the assignation of GLM flashes to LMA, the problem was that two LMA flashes close in time had assigned the same GLM flash. Therefore, those LMA flashes happened at different times and were different flashes. But in this case, the GLM flashes that share a LMA flash also share the time interval, or part of it. It would be interesting to analyse the flashes with this characteristic and discard if it could be possible that the same flash is detected twice by GLM. Table 12. GLM flashes repeated during LMA flash detection and LMA flashes repeated during GLM flash detection. Case GLM flashes repeated LMA flashes repeated A 327 (14.7%) 86 (1.4%) B 0 (0 %) 0 (0 %) C 138 (10%) 58 (1.5%) D 44 (13.5%) 24 (5.5%) Total 509 (12.9%) 168 (1.6%) Table 12 shows the number of GLM flashes that were assigned to more than one LMA flash in section 3.2, and the LMA flashes that were assigned to more than one GLM flash in the current analysis. As expected, the number of repeated flashes is lower when using GLM flashes as reference, as the issue seems to be more like a punctual identification problem. Figure 22 also helps to validate the assumption made for the repetition of flashes in the LMA assignation, since more than 50% of GLM flashes are correlated to various LMA flashes. It would be really interesting to calculate the detection efficiency for GLM groups, in order to compare the number of LMA flashes associated to a group, and see if groups are closer in size to LMA flashes. 3. Results 39 Figure 22. Right: distribution of the number of LMA flashes per GLM flash. Left: Accumulated number of LMA sources per GLM flash. For comparison of parameters between GLM flashes detected and not detected, only the number of events and the radiance of the events is calculated, as GLM does not give information related to the height of the events. Figure 23. Number of GLM events for flashes detected and not detected by LMA. For the number of events per flash, as for LMA sources, it seems that detected flashes have a higher number of events. The median number of events for detected is 111 and for not detected is 86. 3. Results 40 Figure 24. Radiance of GLM events for flashes detected and not detected by LMA. The difference on radiance between detected and not detected flashes is very small and seems to not have an impact on the detection. To check if that is true, the influence of the radiance of the flashes over the detection efficiency has been calculated. Figure 25. Influence of flash radiance (energy) to Detection efficiency. The change of the detection efficiency with radiance is not as big as with LMA height, as it only changes a 20%, but there is still a decrease on the value related with the decrease of the radiance. 3. Results 41 It is also interesting to see that as opposite as height, were most of the flashes had a higher value, for radiance a major part of the flashes are on the lowest levels detected (average radiance is 0.26). 3.4. Flash Duration A general view of the differences in time duration can be obtained calculating the average duration values for GLM and LMA flashes. The duration of those flashes associated during the flash detection efficiency analysis using LMA and GLM as reference can also be calculated to see if there is a difference between their times and those of all the flashes in data. To obtain the average duration of a flash, the time of the last and the first source or event are subtracted. When the time is calculated for all flashes, the average value for the case can be obtained. In the next table are shown the values obtained for each case. Table 13. Mean time duration (in seconds) for LMA and GLM flashes and those flashes associated to a flash for analysis in sections 3.2 and 3.3. Case GLM flashes GLM associated to LMA LMA flashes LMA associated to GLM A 0.3843 0.4195 0.1770 0.2696 B 0.2840 0.2326 0.1097 0.1217 C 0.4199 0.4921 0.1813 0.2230 D 0.3466 0.4029 0.2205 0.28744 All 0.3855 0.4415 0.1808 0.2495 From these results, the average length of a LMA flash is not even a 50% of the length of a GLM flash. Knowing this information, it is quite understandable that GLM flashes end up matching with various LMA flashes. Another remarkable thing is that flashes associated have longer durations than the average flash, it is also a very coherent result since the longer the flash is there are more possibilities that both systems detect it. 3.5. Location accuracy Using the flash assignation from section 3.2., the sources and events belonging to a correlated flash are distributed in a grid of 0.075º lat x 0.075º lon in order to compare their location.