1. INTRODUCTION

In November 2011, to protect critical facility against cloud-to-ground strikes, as part of positive-lightning protection system (LPS), a stand-alone lightning warning system (SLWS), named ALARM by Vaisala Inc., based on intra-cloud discharges (IC), cloud-to-ground flashes (CG), and atmospheric electric field data (EF) was installed at Jin-cheon, National Meteorological Satellite Center (NMSC), 160 m middle sea level (MSL), Republic of Korea.

Previously, the warning performance of the SLWS with introducing various kinds of warning trigger and clear conditions, including the number of flashes for one minute, data combination, type of flashes, atmospheric electric field data (EF) threshold, and dwell time(DT), has been analyzed for eight years^(1,2,3,4,5). As the results of the prior studies, following have been found.

․ As basic warning definitions, the +1.5 ∼ -3 kV/m threshold warning method using EF with five minute dwell time (DT) was the critical hindrance to reducing FAR⁽⁵⁾.

․ Due to probability of detection (POD) decreased steadily for longer lead times for CG, it was not possible for the system to apply more than two second lead time (LT). As found previously, at twenty minute LT, the POD was diminished by zero⁽⁴⁾.

․ When using IC for trigger conditions, the number of flashes for one minute should be located at two because of decreased POD with one, and increased FAR with three. In the Jin-cheon city as the study area, 24% of the first cloud-to-ground strike in the area of concern (AOC) lagged the first IC by one second to sixty four minutes during the summers 2015, and 2016⁽³⁾.

․ To decrease failure to warn (FTW) (to increase POD), the thirty five minute DT as a clear time interval has been applied since 2016. As an inevitable consequence of the decision, total alarm duration (TAD) including LT, warning time (WT), and dwell time (DT) has became longer than before⁽²⁾.

․ The logical OR (union) combination with IC and CG was better than AND (intersection) option of them to POD, whereas the critical success index (CSI) as the function of both POD and false alarm ratio (FAR) became worse than IC only⁽¹⁾.

In this study, to ascertain the way to increase the CSI, the warning performance with logical AND combination between IC and EF has been examined during the summers of 2018, and 2019. Warning statistics with CG in the area of concern (CGAOC), successful prior warning (SUC), false alarm (FR), and FTW have been collected to evaluate the warning conditions. The collected statistics have been analyzed, and compared with other conditions including prior literature, such as Murphy and Said (2016), by using POD, FAR, CSI, and TAD⁽⁶⁾.

2. STAND ALONE LIGHTNING WARNING SYSTEM (SLWS)

The composition, specification, and warning method of the SLWS used in this study are the same as in my past studies^(1,2,3,4,5). To help readers understand better, Table 1, and Fig 1 show the information of the SLWS again.

Fig. 1. Fixed two areas and one area for warning method of the SLWS

When using CG for the type of flashes as warning conditions, the two areas warning method composed of AOC and warning area (WA) is illustrated in Figure 1(left). AOC and WA are circles with radii of 9 and 19 km. The AOC area for which warnings are needed is surrounding the central point of interest (PI). The AOC is surrounded by a second region named the WA. The primary purpose of monitoring lightning activity within the WA is to provide advance notice of the possibility of cloud-to-ground strikes in the AOC. To raise the POD(to decrease FTW), all sectors within WA can be selected. On the other hand, only the three western most sectors for the prior warning tier can be chosen to decrease FA(to increase CSI) since the location of NMSC where thunderstorms tend to move in from the west. Because there is a trade-off between the POD and FAR in the system, it is not possible to optimize both POD and FAR at the same time. Therefore, users introducing the system should decide where to put their goal.

Figure 1(right) shows warning method called one area when using IC. Due to the minimal and omnidirectional detection of IC data from the system, radius is limited at 16 km without sectors. The warning methods adopted in this paper are the one area for IC, and threshold warning for EF. Those are similar to the “lightning AND electric field rule” used by Murphy and Said (2016)⁽⁶⁾.

The SLWS is set up to enable a red light, siren and to turn on/off power of critical systems whenever warning tier triggered and keep the warning going until set DT after the warning condition is below setting value. The Table 2 provides detail on the trigger and clear conditions used by each warning tier. High Alert 1 is a condition in which lightning is already presenting in the AOC, whereas other tiers are for prior warning. According to the results of prior studies^(1,2,3,4,5), each condition, such as the number of flashes for one minute, is chosen to raise CSI. In order to evaluate the behavior of lightning warnings triggered by EF threshold, different threshold values are considered varying between +1.0 and –1.5 kV/m through the Test 1, 2 tier. The only difference of warning conditions between Test tiers is the EF threshold. As noted in the introduction section, the thirty five minute is adopted in this paper as DT.

3. WARNINGS

3.1 Basic Definitions

The three essential metrics of warning statistics, successful warning (SUC), FTW, and FA, are summarized in Table 3. The CG in the area of concern (CGAOC) is the number of the first occurrence CG in AOC. The POD, FAR, CSI, and FTW are calculated and verified each other by following equations. Using the equations, and Table 3 to anticipate warning performance is such a widely used method except for CSI which has been only adopted by my prior studies and here^(1,2,3).

Table 3. Contingency table for warnings

		Triggering Flashes(CG or IC) in AOC (or WA)
		Yes	No
Warning	Yes	SUC	FA
	No	FTW

Figure 2 describes the decisive tree for effective alert (EA), it is used to decide whether one SUC is EA or not. Because one second is enough to turn on/off power automatically in case of the NMSC, it is referred to as effective LT⁽²⁾.

Fig. 2. Decisive tree for Effective Alert

3.3 Comparison with Others

The CSI as a function between POD and FAR is a useful index to compare one results directly with other warning performance. In Table 5, direct performance comparisons with other literatures are made by using CSI, even though making the comparisons are insufficient to confirm the reliability of warning conditions given that there are numerous differences in them^(6,7,8,9). Every POD and FAR in Table 5, is the average values during analysis duration. Due to CSI in other studies associated with lightning warning performance is not founded, the index has been calculated at the table. The best CSI in others is 0.46 by Murphy, and Said (2016) with “EF AND (IC OR CG)”, two minute LT, 50% rate of AOC/WA, and thirty minute DT⁽⁶⁾. In this study, the best CSI is 0.63, 37% increment as against their 0.46. Therefore, the condition used in here with EF AND IC, one second LT, 47% rate of AOC/WA, and thirty five minute DT, is more effective than others.

Table 4. Warning statistics and performance from 2015 to 2019

Year	Warning Tier (data combination)	Statistic				Performance
		CG AOC	SUC	FTW	FA	POD	FAR	CSI	LT(minute)	TAD (minute)
									Average
2019	Test 1(IC AND EF)	17	13	4	3	0.76	0.19	0.65	52	206
	Test 2(IC AND EF)		12	5	4	0.71	0.25	0.57	14	90
	1) warning(CG)		8	9	11	0.47	0.58	0.29	19	92
	2) high alert 2(IC)		13	4	7	0.76	0.35	0.54	35	159
	1) OR 2)(CG OR IC)		14	3	12	0.82	0.46	0.48	43	162
	1) AND 2)(CG AND IC)		7	10	5	0.41	0.42	0.32	16	94
2018	Test 1(IC AND EF)	16	12	4	4	0.75	0.25	0.60	94	242
	Test 2(IC AND EF)		10	6	4	0.63	0.29	0.50	25	123
	1) warning(CG)		7	9	11	0.44	0.61	0.26	45	108
	2) high alert 2(IC)		13	3	7	0.81	0.35	0.57	35	146
	1) OR 2)(CG OR IC)		15	1	16	0.94	0.52	0.47	46	79
	1) AND 2)(CG AND IC)		5	11	4	0.31	0.44	0.25	42	116
2017	watch 1(IC)	25	20	5	14	0.80	0.41	0.51	48	139
	watch 2(IC)		19	6	5	0.76	0.21	0.63	39	137
	1) warning(CG)		10	15	5	0.40	0.33	0.33	33	98
	2) high alert 2(IC)		21	4	8	0.84	0.28	0.64	61	207
	1) OR 2)(CG OR IC)		21	4	9	0.84	0.30	0.62	62	207
	1) AND 2)(CG AND IC)		10	15	3	0.40	0.23	0.36	31	131
2016	watch(IC)	12	10	2	35	0.83	0.78	0.21	129	463
	1) warning(CG)		6	6	2	0.50	0.25	0.43	8	71
	2) high alert 2(IC)		10	2	10	0.83	0.50	0.45	55	471
	1) OR 2)(CG OR IC)		10	2	11	0.83	0.52	0.43	55	471
	1) AND 2)(CG AND IC)		6	12	1	0.50	0.14	0.46	8	68
2015	watch 2(IC)	16	14	2	55	0.88	0.80	0.20	19	106
	warning(CG)		1	15	1	0.06	0.50	0.06	1	120
	high alert 2(CG)		3	13	2	0.19	0.40	0.17	6	98

Fig. 3. Warning performance in POD, FAR, and CS

Fig. 4. Warning performance in LT, and TAD

Table 5. Comparison with other literatures

Reference	Analysis Duration	Effective Lead Time	Data Combination	Radius (㎞)		Rate(%) AOC/WA	Dwell Time (minute)	POD	FAR	CSI
				AOC	WA
Present study (2020) (NMSC, Korea)	2015 ∼ 2019	1 Sec.	only CG	9	19	47	35	0.38	0.45	0.27
	2015 ∼ 2019		only IC	16	N/A	N/A	35	0.82	0.46	0.48
	2016 ∼ 2019		CG OR(+) IC	9	19	47	35	0.85	0.48	0.47
	2016 ∼ 2019		CG AND(×) IC	9	19	47	35	0.41	0.33	0.34
	2018, 2019		EF AND(×) IC	16	N/A	N/A	35	0.76	0.19	0.65
Martin J. Murphy et al. (2016) (US)(6)	2015	2 Min.	EF OR(+) (IC OR CG)	10	20	50	30	1.00	0.78	0.22
			EF AND(×) (IC OR CG)	10	20	50	30	0.71	0.43	0.46
			CG OR(+) IC	10	20	50	30	1.00	0.71	0.29
Holle et al. (2016) (US)(7)	2014	2 Min.	CG OR(+) IC	0.5	5	10	15	0.88	0.89	0.11
				0.5	10	5	15	0.97	0.94	0.06
				0.5	15	3	15	0.97	0.96	0.04
				4.8	15	32	15	0.90	0.72	0.27
Schmitt et al. (2016)(EU)(8)	2010 ∼ 2015	20 Min.	CG OR(+) IC	10	20	50	60	0.82	0.65	0.33
Holle et al. (2014) (US)(9)	2013	2 Min.	CG OR(+) IC	4.8	15	32	15	0.83	0.71	0.27
		20 Min.					15	0.42	0.71	0.21
		2 Min.		4.8	10	48	15	0.69	0.58	0.35

4. CONCLUSIONS

The objective of SLWS, introduced as the positive lightning protection system in 2011, is to trigger warnings before the first cloud-ground strikes around the NMSC with one second lead time. To date, the warning performance of the system has been analyzed by adopting various approaches to raise the performance for eight years^(1,2,3,4,5). In this study, to increase in CSI as the balance between POD and FAR, a combinational logic AND of IC and EF was added with thirty five minute DT for IC, fifteen minute DT for EF, +1.0 ∼ -1.5 kV/m EF threshold, and one area warning method over two summers in 2018, and 2019. The best CSI as the average value in past two years was 0.63, which was resulted from decrease in FAR to 0.22. The index value was better than those of my past studies, and other conditions including other literatures, such as Murphy, and Said (2016). Whereas POD was somewhat became worse to 0.76 than 0.82 of IC alone, and 0.85 of IC OR CG. Taking into account the EF threshold, which was the only difference of warning conditions between test tiers, using the +0.5 ∼ -1.0 kV/m was better than +1.0 ∼ -1.5 for improvement of POD, FAR, and CSI. As the shortest and best TAD of ninety three minutes among the assessed warning conditions, containing LT, WT and DT, occurred at CG alone. However, considering the performance, 0.37 of POD, 0.45 of FAR, and 0.27 of CSI, it is inefficient for a warning condition. The longest 224 minutes in TAD with IC AND EF, was the worst case. But there were very little difference in the durations of the top three CSI and POD conditions, IC only, and IC OR CG. The results suggest that adopting the IC AND EF condition is considerable to augment CSI except for TAD, at least in summer thunderstorms at this study area, Jin-cheon, Korea. In case of the NMSC, since the cost of FTW is greater than FA, high POD is more important than other performance to protect the critical facility of satellite ground station. Therefore, the IC OR CG definition would not be superseded by IC AND EF.