Lightning risk assessment at high spatial resolution at the residential sub-district scale: a case study in the Beijing metropolitan area

ABSTRACT Lightning risk indexes identifying the potential number of dangerous lightning events (NDLE) and ground sensitivity to lightning in residential sub-districts in the Beijing metropolitan area have been estimated on a 5 m resolution grid for the first time. The gridded cloud-to-ground (CG) lightning strike density was used in the NDLE calculation, on account of the multiple contacts formed by CG events with multiple lightning flashes. Meanwhile, in the NDLE estimates, the critical CG strike densities derived from the lightning location system data were corrected for network detection efficiency (DE). The case study for a residential sub-district indicates that the site-specific sensitivity to lightning, which is determined by the terrain factors related to lightning attachment and the lightning rod effects induced by nearby structures, differs greatly among types of underlying ground areas. The discrepancy in the NDLE, which is the numerical product of sensitivity and CG strike density, is dominated by the sensitivity to the relatively stationary CG strike density at the residential sub-district scale. Conclusively, the visualization of lightning risk sensitivity and NDLE differences in parts of a residential sub-district at a high spatial resolution makes this model useful in risk reduction and risk control for lightning risk management.


Introduction
The frequent occurrences of lightning disaster events cause large numbers of casualties and substantial damage losses, such that lightning is considered one of the most dangerous natural hazards (Curran et al. 2000;Holle et al. 2005;Zhang et al. 2011) and the second most fatal meteorological phenomenon (Ashley and Gilson 2009). Lightning risk assessment is meant to investigate and locate high-risk areas, enabling the implementation of mitigation measures for lightning risk reduction (Kaplan and Garrick 1981;Hu et al. 2014). Previously, we developed a lightning risk zoning model based on 1 km resolution grids . The lightning risk recognition at that scale, however, does not appear to be fine enough to reflect the lightning risk characteristics that are useful in disaster preparedness, especially in densely populated urban areas. Thus, it is desirable to assess lightning risk at an extremely high resolution (e.g. a 5 m spacing grid) at the residential sub-district scale in order to identify risk discrepancies in detail, which have significance for risk control and risk reduction (Mills et al 2010).
High spatial resolution enables deliberately locating specific underlying areas, especially in densely populated urban areas. Thus, high resolutions improve the estimates of ground sensitivity to lightning, which is correlated with certain environmental settings, such as topographical features and distribution of earthen structures (Rizk 1994;Vogt 2011). Approaches will be employed in pattern recognition of topographical features, locating earthen structures and determining their lightning collection areas, downscaling grids of cloud-to-ground (CG) strike densities, among others. These processes can be accomplished with the support of GIS technology using high-resolution map data. Natural lightning CG strikes are obvious drivers of lightning-related disasters. Lightning climatology, which preliminarily denotes lightning risk, should be quantified for risk assessment (Bogdan and Burcea 2010). The approach is to derive lightning parameters (e.g. CG flash/strike density and CG flash multiplicity) from observational data, e.g. climatological data (Changnon 1985;Gabriel and Changnon 1989), remote sensing lightning imagery (Christian et al. 2003) and lightning location system (LLS) data (Changnon 1993;Schulz et al. 2005;Biagi et al. 2007;Cummins and Murphy 2009). These lightning parameters fundamentally reflect regional lightning activity relevant to lightning disaster occurrence (Schulz et al. 2005;M€ akel€ a et al. 2010). They are critical in confirming lightning risk even at the residential sub-districts scale.
As a premise of risk recognition, lightning characteristics should be revealed mostly by introducing LLS data, on account of its high spatial-temporal resolution (e.g. Krider et al. 1980). Then, the lightning risk characteristics can be obtained by overlapping the lightning characteristics (CG flash/ strike density) with other risk factors (e.g. sensitivity and exposure) (Hu 2014).
Lightning risk is linked to the combined effects of regional lightning activities and ground sensitivity to lightning. Risk recognition at high resolution can provide visualizations for the decisionmaking in risk management. It facilitates risk-reduction strategies that are practicable in disaster prevention (Smith 1996). For a residential sub-district, the visual lightning risk recognition can provide information in a form that is straightforwardly understandable to local decision and policymakers. Moreover, this site-specific lightning risk is critical to public safety and infrastructure planning ).

Data description
2.1. Lightning location system (LLS) data LLS data collected from 2007-2016 by the ADTD (Advanced TOA and Direction system; TOA denotes time-of-arrival) deployed by the China Meteorology Administration (CMA) were used to derive the CG flash/strike density. These data include time, location, peak current and polarity of CG lightning strikes.
The ADTD consists of more than 301 sensors (as of March 2011) in China (Yao et al. 2012). In Beijing, 9-14 ADTD-1 sensors [improved IMPACT [combined MDF (magnetic direction finding) and TOA] sensors] can detect 1-450 kHz (the very low-frequency band) lightning sources (Figure 1). The ADTD-1 sensors use the combined MDF and TOA method for position retrieval. In this method, if a lightning source is only detected by two ADTD-1 sensors, the algorithm uses one TOA hyperbolic curve and two MDF vectors to retrieve the position. If it is detected by three sensors in a non-duplicate region, the TOA algorithm is used to retrieve the position directly, whereas the TOA is first used to find a duplicate location, then the MDF is used to find the true location. If a lightning event is detected by four or more sensors, a TOA least square method is used to retrieve a more precise position. Thus, the location precision of the lightning source reported by four or more sensors is better than that reported by fewer sensors. In our LLS data, the percentage of lightning sources reported by four or more sensors relative to the total number of detected sources is 66.815%. Meanwhile, the ADTD-observed +CG and -CG lightning peak currents are in the ranges of 0.08-995.9 and 0.258-992.6 kA, respectively ( Figure 1).
The manufacturers claimed that the detection efficiency (DE) of ADTD sensors could be 90% at distances between 300 and 600 km, with a median location accuracy error of 1 km. However, only the flash DE can be 90%, whereas the strike detection efficiency (SDE) is lower. The first strike peak current in a multiple-strike CG flash can be greater than twice its subsequent strike peak current (Rakov and Uman 1990). Thus, the sensors can capture the first larger peak strike but missing the weaker subsequent strike (Rudlosky and Fuelberg 2010). Moreover, some weak CG strikes (including single-strike CG flashes) cannot be detected due to signal attenuation induced by long-distance propagation and terrain factors (Sch€ utte et al. 1988), among other factors. Because the strike number is critical in lightning risk estimates (Bertram and Mayr 2004), we estimated the SDEs of the ADTD in grids (1 km £ 1 km, see Figure 2) and corrected the lightning strike density using the SDE. The SDE estimates approximate those of the U.S. National Lightning Detection Network (NLDN) in 1998, which was reported to be 62% (Idone et al. 1998). Hence, the DE level of the ADTD is equivalent to that of the NLDN, at least in 1998, indicating that considerable improvement remains in terms of network upgrades.

Other data
Digital elevation model (DEM) data were used to identify site-specific lightning attachment capabilities related to topography (Vogt 2011). The 30 m spatial resolution basically meets the requirements for identifying hypsographic features and confirming terrain factors.
Additionally, basic GIS maps with scales of 1:2,000 in urban settings and 1:50,000 in rural settings have been used to measure the lightning collection areas of structures based on the geometric shape and height of the structure, which are readily available in GIS map layers ). The GIS map-layer data-set has a structure-type field that can be used to determine the lightning protection capability of structures.

Methods
The lightning risk index of the potential number of dangerous lightning events (NDLE) can be reserved for lightning risk zoning at the residential sub-district scale. Correlated to regional lightning activity and site-specific sensitivity to lightning, the NDLE, i.e. N x , can generally be estimated as ) where K denotes the coefficient related to the environmental setting; N g denotes the CG lightning strike density (strikes/yr¢km 2 ); and A d denotes the collection area of the lightning strike, mostly determined by site-specific lightning attractiveness variably based on the type of underlying ground area. Because each strike in a multiple-strike CG flash can produce damage losses and/or casualties, it is reasonable to treat N g as the CG strike density (strikes/yr¢km 2 ).  Cummins et al. 1998;Sch€ utte et al. 1988;Naccarato and Pinto 2009). Although DE can be determined more precisely with observations based on live information on lightning occurrences (e. g. video or tower measurements), this approach has only been experimentally utilized to produce localized DE estimates (Saraiva et al. 2010;Visacro et al. 2010;Warner et al. 2013). The methods of DE estimates using theoretical models are more convenient and applicable in comprehensively confirming a network DE. Sch€ utte et al. (1987,1988) introduced the Weibull distribution for estimating the signal strength acceptance levels for sensors, and this method can be used for network DE calculations. Cummins et al. (1998) also combined the peak current cumulative distribution with a signal-propagating model to estimate the absolute flash DE for the NLDN. Naccarato and Pinto (2009) deduced the DEs using the individual DE probability distribution functions of the sensors based on samples of CG strike data detected by a large network while considering different distances from the sensors and specific peak current ranges. We calculated the DEs of the ADTD in grids accounting for the network performance and sensitivity based on the distances and azimuths among the sensors. The CG lightning peak currents were converted to signal strength with arbitrary units (a. u.) using the method of Sch€ utte et al. (1988), which linearly measures the signal strength with signal propagating distance. Sch€ utte et al. (1987) confirmed the Weibull distribution of lightning signal strength.
Thus, methodologically, the signal acceptance of a sensor can be given by where s min and s max are the lower and upper signal threshold, which will be 20 and 600 a. u., respectively; r 0 is the standard distance, which will be 100 km; r is the distance to the sensor; and a, b, and c are the scale, the shape and the location parameter of the Weibull distribution of signal strength, respectively (Sch€ utte et al. 1987(Sch€ utte et al. , 1988. Only two ADTD IMPACT sensors reporting a strike are required to obtain a valid solution. Thus, the DE on a grid cell can be determined as where A i (r i ) denotes the acceptance of one sensor; r i (i = 1, 2, 3, …) is the distance of the ith nearest sensor to the grid cell center and A the grid cell DE of the network.

CG strike density correction
The CG strike density N' g directly derived from LLS data can be corrected for the DE of the ADTD using the following equation: where N g is the corrected CG strike density and D g is the DE of the grid. The CG strike density in the high-resolution (e.g. 5 m) grid could be downscaled from a coarse (e.g. 1 km) grid or derived directly from the LLS data using a kernel density estimator. Usually, when the location error of each CG lightning strike observed by LLS is given, the probabilistic computational methods based on confidence ellipse are recommendable for deriving CG strike densities in high spatial resolution (Bourscheidt et al. 2014;Etherington and Perry 2017). However, the ADTD doesn't provide the location error of single CG lightning strike. We are obliged to use a bivariate (XY) kernel density method to estimate the CG lightning strike density at the 5 m resolution, which the kernel is assumed to be Gaussian. The ArcGIS software provides the tool of kernel density estimator (Refer to http://desktop.arcgis.com/en/arcmap/10.3/tools/spatial-analyst-toolbox/ kernel-density.htm). In the parameterization of kernel density function, the cell-size is set to 5 m, and the search radius is 1000 m, which corresponds to the magnitude of the ADTD median location accuracy error.
Supplementarily, the approach of inverse distance weighting (IDW) are also used to interpolate the CG strike density from the larger grid (with 1 km spacing), in which 9 grid cells were involved, i. e. the center, up, down, upper left, upper right, left, right, lower left, and lower right 1 km £ 1 km grid cells. Mathematically, the interpolation can be described as where N g5m is the interpolated CG strike density in a 5 m spaced grid cell; n (n 9) is the number of strike-containing cells and the approximately 1 km spaced grid cells; N g (i) is the CG strike density of the ith 1 km spaced grid cell; and r(i) is the distance of the central point of the 5 m grid cell to that of the ith 1 km spaced grid cell.

NDLE estimates for the 5 m spacing grids
We calculated the NDLE for an earthen structure, an outdoor area under a structure canopy (OAUSC) and an open-field area (OFA). Because these types of underlying ground areas differ in lightning protection capability, lightning attractiveness, and lightning attachment induced by top terrains, different methodologies were used to estimate the NDLE. Additionally, the approaches were adjusted based on the conditions of the grids intersecting with these underlying areas ( Figure 3). The spatial relationship of the underlying areas to the grids means that one grid box often covers only one unique underlying area and seldom covers multiple types of areas ( Figure 3). Thus, the NDLE of each grid cell can be calculated directly using a GIS overlapping operator. This approach is different from that used for the 1 £ 1 km sized grids, which is to sum the total NDLE of all areas of the large grid ).

NDLE estimates of an earthen structure (ES)
The NDLE value of a structure N d is calculated as follows ): where A d (m 2 ) is the lightning collection area of a structure; C d is the terrain factor, which is deduced using DEM data, accounting for its relationship to the surrounding topography (see Table 1); and P d is the coefficient representing the lightning protection capability of the structure. Given the structure height in metres H, the collection area A d can be determined as follows (Rizk 1994): The structure protection capability includes protecting 1) living beings from being injured by a lightning strike, 2) the structure from physical damage, and 3) the internal systems in the structure. Substantially, these capabilities are represented by the casualty probability p a , the physical damage probability p b , and the internal systems failure probability p c in risk estimates. Herein, for simplification, the lightning risk assessment only takes into account p a , i.e. P d = p a.
The casualty probability due to touch and step voltage induced by lightning striking the structure reflects the structure's Lightning Protection Level (LPL), which can be deduced by accounting for the lightning protection measures taken by the structure (Table 2).
We listed the protection measures that would be probably taken by 10 structure types in Beijing (see Table 2) available in a GIS map-layer data-set. Most structures are equipped with lightning rods. Some concrete steel structures have iron infrastructure and framework as the lead-in wire for  Table 1. Estimating the terrain factor of a structure based on its surrounding topography (defined by IEC62305-2 2010).
Description of the surrounding topography C d Higher than the top of the structure 0.25 As high as the top of the structure 0.5 On flat ground 1 On top of a hill 2 lightning protection. Thus, they possess a better capability of protecting living beings from lightning strike injuries.

NDLE estimates of an outdoor area under a structure canopy (OAUSC)
Under these conditions, the NDLE value, i.e. N Dc , can be calculated as follows: where A Dc (m 2 ) is the intersection area of the OAUSC and the grid cell; C d is the terrain factor of the grid cell; and C c is the coefficient representing lightning rod effects produced by the surrounding structures. At a fine grid scale (e.g. 5 m), its calculation is simplified as follows (Petrov and D' Alessandro 2002): where H(1),…, H(n) are the floor numbers of the surrounding structures, whose canopies cover the grid cell. It is reasonable that C c will approximate zero if the grid cell is under canopies of many nearby tall structures.

NDLE estimates of an open-field area (OFA)
Because they are totally exposed to lightning strikes, OFAs are more susceptible to lightning. Thus, the NDLE, N Ds , can be estimated as follows: where A Ds is the intersection area of the OFA and the grid cell.

NDLE estimates in a grid cell
After the NDLE values of the three types of underlying ground areas are calculated, the NDLE of a grid cell intersecting with these areas, Nd_Cell, can be calculated as follows:  where Area_Cell, Area_ES, Area_OAUSC, and Area_OFA denote the geometries of the grid cell, the earthen structure, the outdoor area under structure canopy, and the OFA in the grid cell, respectively. Intersect is a GIS operator for calculating the intersection areas of the grid cell and the geometries of the three types of underlying ground areas (i.e. the structure, the outdoor area under structure canopy, and the OFA), respectively.

Parameters reflecting lightning risk characteristics
The lightning risk assessment is to estimate the NDLE and sensitivity and to subsequently identify high-risk areas. Then, pertinent advice can be given to decision-makers who will undertake measures for lightning risk mitigation in residential sub-districts. The CG strike density, N g , ground sensitivity to lightning, S x , and the NDLE, N d , essentially reflect the lightning risk characteristics of a local community, which are critical for decision-making in lightning risk management. The CG strike density, N g , an indicator of regional lightning activity, can be derived from the LLS data. The NDLE value, a numerical product of the CG strike density, N g , and sensitivity, S x , reflects the site-specific lightning hazards.
The sensitivity is defined as an indicator of the lightning strike susceptibility of the underlying ground and is mostly correlated to land-surface characteristics, e.g. terrain features and distribution of earthen structures. Based on site-specific environmental settings rather than regional lightning activity, the sensitivity can be calculated as follows: where S d , S Dc , and S Ds are the lightning sensitivity for a structure, an OAUSC and an OFA, respectively. Differences in the NDLE in a sub-district are mostly determined by the sensitivity, due to the relative constant values of CG strike density. In this context, the sensitivity and the NDLE jointly indicate the lightning risk at a high resolution.

Analysis on lightning characteristics
Lightning climatology preliminarily reflects lightning risk characteristics but does not account for sensitivity and exposure to lightning (Ashley and Gilson 2009). Analysis of lightning characteristics is the premise of risk assessment even at the sub-district scale, as it can provide critical parameters for lightning risk assessment, e.g. the CG flash/strike density and CG multiplicity. We derived the lightning parameters from the ADTD data by counting the annual CG flash/strike numbers in 1 km grids. The CG strikes were grouped into flashes based on a multiplicity delay of 1 s within a radius of 20 km (Cummins et al. 2006;Dr€ ue et al. 2007), and +CG flashes with a peak current of less than 15 kA were classified as IC lightning (recommended by Cummins and Murphy 2009). Convection events are usually enhanced by orographic uplift in the mountains, which trigger more CG strikes (Bourscheidt et al. 2009). However, the derivation from the ADTD data exhibits relatively lower CG flash/strike densities in the northern and western mountainous areas than in the plains, except for a relatively high density in the south-western mountains (Figure 4). The thunderstorms in urban areas on the plains can be enhanced by urban characteristics (e.g. roughness, aerosols, and urban heat islands) and consequently produce more CG flashes (Shepherd et al. 2002;Rose et al. 2008;Stallins and Rose 2008;Hu et al 2014;Hu 2015;Kar and Liou 2014), especially in the downwind areas (see the blue-circled in Figure 4(c)). Additionally, a high CG strike density distributed in upwind southern areas (see the purple-circled in Figure 4(c)) is observed. We assumed that Figure 4. Distribution of (a) CG flash density (flash/yr¢km 2 ), (b) CG strike density (strikes/yr¢km 2 ), and (c) corrected CG strike density (strikes/yr¢km 2 ). For convenience, the same legends for contours and shading were used in the CG flash density, CG strike density and corrected CG strike density plots. (This figure is available in colour online.) this is related to random cloud condensation nuclei concentrations affecting cloud properties and the initiation of precipitation over cities (Steiger et al. 2002;Stallins et al. 2006;Kar and Liou 2014).
No matter what can explains the higher CG flash/strike density in the plains, the DE of an LLS cannot be 100% (Schulz et al. 2005;Mazarakis et al. 2008). The actual CG strike numbers in the grids, however, are critical to the NDLE estimation. Thus, we corrected the gridded CG strike densities for DEs to fit the actual values.
After being corrected using the deduced DEs (see Figure 2), the CG strike densities in the northeastern mountains, metropolitan areas, southern plains and south-western mountains increased significantly in comparison with the uncorrected values (see Figure 4(b) and 4(c)). The corrected densities in metropolitan areas are mainly between 2 and 4 strikes/yr¢km 2 , which are higher than expected. However, a relatively high CG strike density remains in the plains.
Based on samples of LLS data, the probability distribution of lightning signal strength is critical to deducing the network DE. The LLS data observed in a limited period of time cannot provide a perfect probability distribution, which obviously can lead to biases in DE estimates. Moreover, the underlying surface conductivity is lower in mountains, and the lightning signal strength will be more attenuated in mountains than in plains. However, this attenuation cannot be precisely taken into account in DE estimates and, thus, leads to uncertainty. No matter how the DE estimates are effectively used for correcting CG lightning density derived from LLS data, it is still advisable that the network should be upgraded to improve the ADTD DEs and the detection accuracy. The anomaly of higher lightning density in the plains may be explained with additional evidence.

Case study of lightning risk assessment in a residential sub-district
The model running at a 5 m resolution can optimally cover a small area of 10-100 km 2 . We selected two residential sub-districts in Beijing metropolitan areas for risk analysis, accounting for indicators of sensitivity and the NDLE. One is the sub-district of Malianwa in the north-western metropolitan areas and the foothills of the western YanShan Range. Its complex topography involves a diversity of ground sensitivities to lightning. The other is the Beijing International Airport, where the lightning risk discrepancy between the open fields of the aircraft parking areas and the terminal structure is remarkable.

Ground sensitivity to lightning
In terms of risk management, sensitivity recognition contributes to lightning risk avoidance on thunderstorm days. Additionally, it can be used to direct deployment of lightning protection facilities and systems (Schulz et al. 2005).
The lightning sensitivity zoning in the sub-district of Malianwa indicates that the sensitivity magnitudes of structures and outdoor areas under structure canopies are usually less than 0.15 ( Figure 5  (a)). Alternatively, if the terrain factors are not included, the greatest sensitivity is 1.0 on an open field in the plains (Figure 5(a)). Accounting for the terrain factors, the sensitivity in mountainous areas will increase to 1.15-1.3, for example, in the high sensitivity zones of western uplands of this sub-district (see A in Figure 5(b) and 5(c)). This higher sensitivity in the hills means that the CG strikes are more likely to occur at topographic highpoints by as much as 15%-30.0% when compared with random points in the plains. This increased sensitivity of topographic highpoints is somewhat in agreement with the findings of Vogt (2011).
As displayed in Google Earth, the sensitivity zones exhibit a good correlation with topographical features and the distribution of earthen structures (see Figure 5(c)). Especially, on account of the protection from the structures, the structure occupying areas as well as these under the structure canopies exhibit an abrupt lower sensitivity to lighting. A sensitivity buffer can be recognized between the earthed structure and its surrounding open field, where it forms a ring pattern of higher sensitivity values around these structures (see Figures 5 and 6(a)). Apparently, the simulated sensitivity is explicably in accordance with the settings, and it is valuable in visualizing lightning risk management.

NDLE
Similar to sensitivity, the NDLE values for a structure and an OAUSC are lower. The NDLE values for an OFA are equal or even magnitudes greater than the CG strike densities of the downscaled grids. The NDLE values for the uplands in western Malianwa exhibit this pattern, where more upward and/or downward lightning can be triggered by topographic highpoint attachment (Warner et al. 2013). Figure 5. Sensitivity zones in the sub-district of Malianwa, in cases of (a) not accounting for the terrain factor, (b) accounting for terrain factors, and (c) displayed in Google Earth. These zones correspond well with the distribution of underlying structures and topographical features. For example, point A in the mountainous areas exhibits a high sensitivity, B in the dense structure area exhibits a lower sensitivity on account of the lightning rod effects produced by nearby structures, and C in an OFA exhibits a relatively high sensitivity. Interestingly, the sensitivity of point D at an open sports field is obviously higher than that of the surrounding densely built-up areas. (This figure is available in colour online.) Figure 6. Lightning risk assessment of (a) ground sensitivity to lightning, (b) NDLE deduced using corrected CG strike density interpolated by IDW, and (c) NDLE deduced using corrected CG strike density estimated by kernel density method, for the Beijing International Airport. Obviously, the NDLE in (c) appears more smooth than that in (b), all displayed in Google Earth. (This figure is available in colour online.) The advantage of quantitative risk assessment at high resolution is that its visualized risk characteristics can play an important role in operating risk control effectively. For instance, at the Beijing International Airport, terminal 3 (a 45 m high structure) and its nearby outdoor areas under structure canopies exhibit a low assessed sensitivity of 0.15, equivalent to 0.15 times that of an OFA, and NDLE values below 0.2 times/yr¢km 2 ( Figure 6). Conversely, the red ellipse in the aircraft parking apron, hundreds of metres away from the terminal, exhibits a high sensitivity of 1.0, and NDLE values above 1.0 time/yr¢km 2 (in Figure 6(b)), or 2.0 time/yr¢km 2 (in Figure 6(c)), due to the lack of lightning protection and structure shelter. On 11 August 2013, a lightning fatality occurred within the red ellipse (Figure 6(b)), when a cleaning staff member was struck dead by lightning while using a mobile phone (Hu 2014). Therefore, the personnel should pay attention to lightning on thunderstorm days when operating in open fields. Moreover, lightning risk management should be conducted based upon risk recognition in the airport community so that it can visually inform personnel regarding safe and unsafe areas ( Figure 6).

Conclusion
The DE of a LLS cannot be 100%, and low DEs are usually due to a lack of deployed network sensors, as well as the performance and sensitivity of the sensors. Meanwhile, the signals produced by CG flashes can be strongly attenuated by long-distance propagation, terrain factors and underlying land surface conductivity. Before being used in NDLE estimates, the CG strike densities derived from LLS data should be corrected for DEs. Although the correction of CG strike density makes it better qualified for risk assessment, the LLS data should be made more reliable through network upgrades, which can improve the DE and location accuracy (Rudlosky and Fuelberg 2010). Moreover, network upgrades should be implemented not only for optimal lightning locating in metropolitan areas but also in mountainous rural areas, where more lightning casualties occur (L opez and Holle 1998; Curran et al. 2000;Zhang et al. 2011).
Uncertainty in lightning risk estimates at this high resolution is influenced by the LLS data quality related to locational precision and imperfect network DEs. Additionally, the model structures and operations (e.g. CG strike density downscaling) can magnify the uncertainty. Although the IDW interpolation and the overlapping of the derived CG strike density with the ground sensitivity to lightning may attenuate the errors in the risk estimate, uncertainty remains. However, it is suggested that the uncertainty caused by the LLS data quality can be reduced through network upgrades by adding high-performance and highly sensitive sensors. Further research can be undertaken to evaluate the reliability of the risk estimates in terms of the uncertainty (e.g. Monte Carlo simulation). Additionally, finding an effective approach for uncertainty reduction is also critical to identifying a more precise calibration and correction process in lightning risk assessment.
The model running at a fine resolution (e.g. a 5 m grid) can be used to accessibly assess lightning risk in terms of ground sensitivity for different types of underlying ground areas, and the data can be overlapped with CG strike density data. The lightning risk recognition at high resolutions can visually reveal risk discrepancies and indicate higher risk areas at a finer scale, making it favourable in lightning risk management.
This case study indicates that the lightning rod effects of structures produce outdoor areas of low risk under its canopy. In comparison, an OFA usually exhibits a higher risk, with an NLDE equal to the corresponding CG strike density and a sensitivity of nearly 1.0 in magnitude. The NLDE and sensitivity can differ by 1.15-1.3 times between uplands and the plains due to higher lightning attachment in elevated areas.
The distributions of lightning parameters (e.g. CG flash/strike density), ground sensitivity to lightning and NDLE comprehensively reveal lightning risk characteristics. The CG lightning flash/ strike density, CG flash multiplicity, and other factors derived from LLS data not only indicate the regional lightning activity but also constitute the input parameters for lightning risk assessments. The sensitivity is correlated to the site-specific lightning protection capability the lightning attractiveness of an earthen structure, and the lightning attachment induced by top terrains. This parameter indicates which parts of a residential sub-district are relatively prone to lightning strikes. The NDLE reflects lightning hazards, accounting for both regional lightning activity and sensitivity. The CG strike density, sensitivity and NLDE are practical indicators for decision-making in lightning risk management. They play important roles when taking effective actions to reduce site-specific lightning risks in residential sub-districts, e.g. erecting warning boards in high-risk areas, installing lightning protection facilities in the domains susceptible to lightning, and even constructing temporary structures serving as thunderstorm shelters in public OFAs.