Susceptibility assessment of wildfire-induced transmission line tripping using a physical-Bayesian modeling approach Authors: Hui Huang, Keyu Chen, Biao Song, Changmeng Chen, Linlin Li, Jianli Ling Published: 2025 ABSTRACT The escalating frequency and intensity of wildfires, driven by extreme climate events and anthropogenic activities, pose severe threats to the security and reliability of electrical power systems. Wildfires in proximity to high-voltage transmission lines can lead to a critical reduction in air insulation strength, caused by elevated temperatures, smoke particulate matter, and highly conductive flames. These conditions often precipitate insulation breakdown, line tripping, and potentially widespread power outages. This paper proposes an integrated physical-Bayesian modeling framework for assessing the susceptibility of transmission lines to wildfire-induced tripping. The framework combines a Bayesian network model, which estimates wildfire ignition probability by incorporating geographic, meteorological, fuel, and anthropogenic variables, with a physics-based model that evaluates the risk of insulation failure under wildfire conditions, accounting for both flame bridging and smoke effects. Key influencing features are identified through the Relief algorithm for feature selection. Applied to the Guangdong Power Grid in China, the model identifies distance to settlements, vegetation type, and monthly rainfall as the dominant factors affecting ignition probability. Spatial analysis reveals strong agreement between high-risk zones and historical fire events, and predicted tripping risks are consistent with actual outage records. The approach supports differentiated wildfire prevention strategies and enhances grid resilience, offering a scientifically robust and practical tool for proactive risk management in power systems exposed to wildfire hazards. Introduction Global climate change has precipitated a surge in the frequency and severity of extreme weather events, among which wildfires have exhibited a pronounced upward trend in occurrence, duration, and spatial extent1–5. These fires inflict considerable ecological and economic damage and present a critical threat to essential infrastructure, particularly electric power systems6–9. Transmission lines, often traversing vast and remote vegetated areas, are highly vulnerable to wildfires. When fires approach transmission corridors, the combined effects of intense heat, smoke, and ionized gases substantially degrade the dielectric strength of the air insulation around conductors. Elevated temperatures reduce air density, smoke particles distort local electric fields, and flames—which are inherently conductive—can bridge air gaps, collectively increasing the probability of flashover and line tripping10–12. Such events can trigger cascading failures, leading to extensive blackouts with profound socio-economic repercussions13–15,. Consequently, the development of robust methodologies for assessing and mapping the susceptibility of transmission lines to wildfire-induced tripping is imperative for enhancing the resilience and operational security of power grids. Existing research on wildfire-related transmission line failures has evolved along several trajectories, including fire spread modeling16,17, characterization of the electrical properties of flames and smoke18,19, and experimental investigation of insulation breakdown mechanisms20,21. For predicting fire behavior, semi-empirical and semi-physical models, such as the widely adopted Rothermel22 model, have been extensively applied23–25. These models integrate fuel properties, environmental conditions, and topographic features to estimate fireline intensity and rate of spread, forming the cornerstone of systems like the U.S. National Fire Danger Rating System (NFDRS) and the BEHAVE fire prediction toolkit26,27. Such models provide critical inputs for operational fire management but are often limited to fire propagation dynamics without explicit linkage to power system vulnerabilities. In the realm of electrical engineering, significant experimental and simulation efforts have been devoted to understanding how flames and smoke compromise air insulation12,28. Studies have demonstrated that combustion products from vegetation- such as charged particles, ash, and ionized gases- can reduce the breakdown voltage of an air gap to less than 40% of its value under clean air conditions29,30. Factors including particle concentration, size distribution, and local temperature gradients have been shown to critically influence discharge initiation and propagation31,32. Furthermore, the phenomenon of flame bridging, where conductive flames physically connect energized conductors to ground or other phases, has been identified as a particularly acute risk factor33,34. Despite these advances, prevailing approaches to assessing wildfire-induced tripping risk often exhibit notable limitations. Many studies focus on isolated aspects- such as estimating ignition frequency from historical fire data or evaluating breakdown probability based solely on flame height- without integrating the full spectrum of causal factors30,35. This neglects the complex interdependencies among ignition likelihood, fire behavior, and the physical mechanisms of insulation failure. Moreover, the spatial heterogeneity of risk, influenced by varying geographical, meteorological, combustible, and anthropogenic conditions, is frequently oversimplified. Many existing models also lack sufficient granularity to support differentiated grid planning and targeted mitigation measures36,37. To address these gaps, this paper introduces a comprehensive integrated framework for assessing the susceptibility of transmission lines to wildfire-induced tripping. The approach combines a Bayesian network model for predicting spatial wildfire occurrence probability with a physics-based model for evaluating insulation breakdown risk under fire conditions. The Bayesian model incorporates multi-source data—spanning geographic, meteorological, fuel, and anthropogenic factors—and employs the Relief algorithm for feature selection to identify dominant predictors. The physical model accounts for both flame bridging and smoke effects on withstand strength, enabling a scenario-based evaluation of breakdown risk. By coupling these components, the framework allows for a spatially explicit and mechanistically grounded calculation of tripping risk along transmission corridors. Unlike previous hybrid models that often focus solely on fire propagation or electrical breakdown in isolation, our framework integrates spatial ignition probability with dynamic physical modeling of insulation failure, enabling an end-to-end risk assessment that is both spatially explicit and mechanistically grounded. Physical model for wildfire-induced transmission line tripping Wildfire spread model Accurately predicting wildfire spread through purely physical mechanistic models remains challenging. Predominant approaches therefore blend physical principles with empirical analysis, yielding semi-physical or semi-empirical models38–40. These models incorporate assumptions regarding heat flux and ignition energy requirements without rigorously delineating heat transfer modes or absorption mechanisms- key constants governing heat flux and absorption are typically derived from experimental data. Furthermore, they do not explicitly model the processes governing net energy transfer per unit area on the fuel bed surface, which is often estimated empirically as a function of reaction intensity. Among these, the model proposed by Rothermel22 has garnered widespread acceptance due to its validation through extensive laboratory experiments and field observations41–44. Table 1Classification of input variables for the Rothermel22 fire spread model.Fuel particle propertiesFuel array configurationEnvironmental parametersHeat content (kJ/kg)Fuel load (kg/m²)Wind speed (m/s)Mineral content (%)Average particle size (mm)Fuel moisture content (%)Particle density (kg/m³)Average fuel bed depth (m)Slope (degrees) Table 1Classification of input variables for the Rothermel22 fire spread model.Fuel particle propertiesFuel array configurationEnvironmental parametersHeat content (kJ/kg)Fuel load (kg/m²)Wind speed (m/s)Mineral content (%)Average particle size (mm)Fuel moisture content (%)Particle density (kg/m³)Average fuel bed depth (m)Slope (degrees) Classification of input variables for the Rothermel22 fire spread model. The Rothermel model categorizes the requisite input variables for predicting wildfire spread into three primary classes, as detailed in Table 1. The spread rate is expressed by the following equation:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R=\frac{{{I_{\mathrm{R}}}\xi (1+{\phi _{\mathrm{w}}}+{\phi _{\mathrm{s}}})}}{{{\rho _{\mathrm{b}}}\varepsilon {Q_{{\mathrm{ig}}}}}}$$\end{document} Where R denotes the forward rate of spread of the fire front (m/min); IR signifies the reaction intensity (kJ/m²·min); ξ represents the propagating flux ratio (dimensionless); \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\phi _{\mathrm{w}}}$$\end{document}is the wind coefficient (dimensionless); \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\phi _{\mathrm{s}}}$$\end{document} is the slope factor (dimensionless); \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho _{\mathrm{b}}}$$\end{document} indicates the fuel bed bulk density (kg/m³); ε is the effective heating number (dimensionless); and Qig is the heat of pre-ignition for the fuel (kJ/kg). The numerator in this expression corresponds to the heat flux reaching the unburned fuel, while the denominator represents the energy required to raise the fuel to its ignition temperature. Fireline intensity is a crucial parameter for establishing operational limits in prescribed burning and fire suppression tactics, as well as for estimating necessary firebreak widths. It also correlates effectively with observable flame characteristics, such as flame length45. Byram46 introduced a seminal model for fireline intensity, defining it as a function of heat yield per unit mass of fuel, available fuel consumption, and rate of spread:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I=HWR$$\end{document} Here, I is the fireline intensity (kW/m); H is the heat yield per unit mass of fuel (J/g); and W is the mass of fuel consumed per unit area (g/m²). (Note: Original unit for W was grams (g), but typically expressed per unit area in fire science). Subsequently, flame height can be estimated using the following relation:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H_{\mathrm{f}}}={\left( {\frac{I}{{250}}} \right)^{1/2}}={\left[ {\frac{{HW{I_{\mathrm{R}}}\xi (1+{\phi _{\mathrm{w}}}+{\phi _{\mathrm{s}}})}}{{250{\rho _{\mathrm{b}}}\varepsilon {Q_{{\mathrm{ig}}}}}}} \right]^{1/2}}$$\end{document} Equation (3) is derived from Heskestad47 and is widely used in fire behavior modeling33,34. As a combination of the Rothermel model, Byram model, and Heskestad model, Eq. (3) effectively establish the relationship between fire spread factors (i.e., fuel particle properties, fuel array configuration, environmental parameters) and flame height. Calculation of average smoke withstand field strength The average smoke withstand field strength under wildfire conditions refers to the maximum electric field intensity that a high-voltage transmission line can endure within a smoke-laden environment caused by a spreading fire without experiencing flashover. The intense heat from vegetation combustion reduces air density and humidity, facilitating the development of electron avalanches and streamers. Additionally, ash and smoke particles, transported by electric field forces and thermal buoyancy, infiltrate the air gaps around conductors, causing electric field distortion and potentially forming conductive pathways, thereby degrading the gap’s withstand capability29,30. Consequently, beyond accounting for variations in air density and humidity, a particle correction factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${K_{\mathrm{p}}}$$\end{document} is introduced to adjust the average withstand field strength applicable to a clean air gap:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E_{\mathrm{s}}}={K_{\mathrm{p}}}{K_{{{\upsigma}}}}{K_{\mathrm{h}}}{E_{\mathrm{a}}}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${K_{\mathrm{\boldsymbol{\upsigma}}}}$$\end{document}and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${K_{\mathrm{h}}}$$\end{document} are the air density and air humidity correction coefficients, respectively; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E_{\mathrm{a}}}$$\end{document} is the average withstand field strength of a pure air gap under standard atmospheric conditions (temperature 20 °C, humidity 11 g/m³, atmospheric pressure = 101.3 kPa). The propensity of ash/smoke particles to initiate discharge within the flame plume depends on factors like particle size, concentration, and local temperature. Experimental evidence29,30 indicates that even smoke from grass combustion can reduce the breakdown voltage of an air gap to below 40% of its value in pure air. Thus, the particle correction factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${K_{\mathrm{p}}}$$\end{document} is conservatively assigned a value of 0.4. The air density correction coefficient \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${K_\sigma }$$\end{document} is proportional to the relative air density \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma$$\end{document} within the smoke environment, raised to a power m:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${K_\sigma }={\sigma ^m}$$\end{document} The relative air density \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma$$\end{document} is calculated as:6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma =\frac{{(273+{T_{\mathrm{a}}}){P_{\mathrm{f}}}}}{{(273+{T_{\mathrm{a}}}+\Delta T){P_{\mathrm{a}}}}}$$\end{document} where m is an air density correction exponent, dependent on gap length, simplified here to m = 1; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${T_{\mathrm{a}}}$$\end{document} is the ambient temperature (°C); \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${P_{\mathrm{f}}}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${P_{\mathrm{a}}}$$\end{document} denote flame zone pressure and ambient pressure, respectively. Given the minimal pressure variation expected at the fire site, it is assumed \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${P_{\mathrm{f}}} \approx {P_{\mathrm{a}}}$$\end{document}; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta T$$\end{document}represents the temperature rise due to combustion, estimable from flame modeling:7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta T=\frac{{3.9\sqrt[3]{{{I^2}}}}}{{{H_{\mathrm{s}}}}}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H_{\mathrm{s}}}$$\end{document} is the height above the vegetation bed to the midpoint of the smoke gap of interest.8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H_s} = \left\{ {\begin{array}{ll} {{d_{{\mathrm{lt}}}}} & \quad {{\mathrm{for}}\;{\mathrm{phase-to-phase}}} \\ {{H_f} + \left( {{d_{{\mathrm{lt}}}} - {H_f}} \right)/2} & \quad {{\mathrm{for}}\;{\mathrm{conductor-to-ground}}\;{\mathrm{gap}}.} \end{array}} \right.$$\end{document} The air humidity correction coefficient \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${K_{\mathrm{h}}}$$\end{document} is given by:9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${K_{\mathrm{h}}}={k^w}$$\end{document} where w is the humidity correction exponent, related to gap distance; k is the humidity correction base, a function of absolute humidity h and the relative air density σ within the smoke:10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k=1+0.012 \times (h{\mathrm{/}}s - 11)$$\end{document} For short air gaps, breakdown voltage exhibits a nearly linear relationship with gap length. However, for longer gaps, the breakdown mechanism transitions from streamer to leader discharge, introducing a saturation effect where the breakdown voltage increases sub-linearly with length, consequently reducing the average breakdown field strength. Based on the Rizk model48,49, for air gap length \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d_s}<4{\kern 1pt} {\kern 1pt} {\mathrm{m}}$$\end{document}, the average withstand field strength \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E_{\mathrm{a}}}$$\end{document} is approximately constant: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E_{\mathrm{a}}} \approx 241.39{\kern 1pt} {\kern 1pt} {\mathrm{kV/m}}$$\end{document}. For air gap length \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d_s} \geqslant 4{\kern 1pt} {\kern 1pt} {\mathrm{m}}$$\end{document}, the average withstand field strength \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E_{\mathrm{a}}}$$\end{document} is determined by:11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E_{\mathrm{a}}}=\frac{{1850+59{d_{\mathrm{s}}}}}{{{d_{\mathrm{s}}}+3.89}}+\frac{{92}}{{{d_{\mathrm{s}}}}}$$\end{document} Note: Equation (11) yields the withstand voltage for the gap. To obtain the average withstand field strength, this result must be divided by the gap length. In summary, the average smoke withstand field strength \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E_{\mathrm{s}}}$$\end{document} under wildfire conditions can be expressed as:12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E_{\mathrm{s}}}=\left[ {0.3472{{\left( {1+\frac{{3.9\sqrt[3]{{{I^2}}}}}{{273+{T_{\mathrm{a}}}}}} \right)}^{ - 1}}+0.0048h} \right] \times \left( {\frac{{1850+59ds}}{{ds+3.89}}+\frac{{92}}{{ds}}} \right)$$\end{document} Transmission line breakdown evaluation Wildfire events within a transmission corridor can precipitate two primary breakdown modes: conductor-to-ground and phase-to-phase. Thus, the overall insulation breakdown risk \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R_{{\mathrm{tf}}}}$$\end{document}under wildfire conditions is defined as the maximum of the conductor-to-ground risk \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R_{\mathrm{g}}}$$\end{document} and the phase-to-phase risk \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R_{\mathrm{p}}}$$\end{document}:13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R_{{\mathrm{tf}}}}=\hbox{max} ({R_g},{\kern 1pt} {\kern 1pt} {R_p})$$\end{document} The individual risks are related to the respective withstand voltages and the system line voltage:14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R_{\mathrm{g}}}=\frac{{\sqrt {2/3} U}}{{{U_{\mathrm{g}}}}}$$\end{document}15\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R_{\mathrm{p}}}=\frac{{\sqrt 2 U}}{{{U_{\mathrm{p}}}}}$$\end{document} where U is the RMS line-to-line voltage of the transmission line; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U_{\mathrm{g}}}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U_{\mathrm{p}}}$$\end{document} represent the calculated withstand voltages (RMS or equivalent) for the conductor-to-ground and phase-to-phase air gaps, respectively, under wildfire conditions. Fig. 1Schematic diagram of the relationship between flame and transmission line position: (a) Flame fully bridging the conductor gap; (b) Flame partially bridging the conductor gap. Schematic diagram of the relationship between flame and transmission line position: (a) Flame fully bridging the conductor gap; (b) Flame partially bridging the conductor gap. The withstand voltage of an air gap during a wildfire is highly dependent on the combustion characteristics. Firstly, chemical dissociation and thermal ionization of alkali and alkaline earth metals in vegetation generate significant charge carriers, conferring high electrical conductivity to the flame itself. If the flame bridges a significant portion of the gap, breakdown is highly probable. Secondly, the presence of burning particles and the heated, smoke-laden atmosphere above the flame significantly reduce the breakdown field strength compared to clean air. Therefore, calculating Ug and Up necessitates considering two distinct scenarios based on whether the flame fully bridges the gap, which depends on the flame height relative to conductor and vegetation heights.Flame Fully Bridging the Conductor Gap.If the vegetation flame height exceeds the conductor height (considering sag), as depicted in Fig. 1(a), both the conductor-to-ground and phase-to-phase insulation paths are assumed to be fully immersed in the flame. Given the rapid carbonization of burning wood and its resulting high electrical conductivity compared to air, the flaming vegetation is approximated as a conductor. Consequently, the withstand voltages Ug and Up are determined primarily by the electrical strength of the flame itself.16\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U_{\mathrm{g}}}=({H_{\mathrm{l}}} - {H_{\mathrm{t}}}){E_{\mathrm{f}}}={d_{{\mathrm{lt}}}}{E_{\mathrm{f}}}$$\end{document}17\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U_{\mathrm{p}}}={d_{\mathrm{p}}}{E_{\mathrm{f}}}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H_{\mathrm{l}}}$$\end{document} is the conductor height above ground, accounting for sag; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H_{\mathrm{t}}}$$\end{document}is the maximum canopy height beneath the line; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d_{{\mathrm{lt}}}}$$\end{document} is the length of the air gap between the conductor and the canopy top; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d_{\mathrm{p}}}$$\end{document} is the phase-to-phase distance; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E_{\mathrm{f}}}$$\end{document} is the average withstand field strength of the flame. Note that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d_{{\mathrm{lt}}}}$$\end{document} represents the vertical height in the present study. While flame tilt due to wind is not explicitly modeled here, the conservative assumption of vertical flame height ensures that the worst-case scenario for gap bridging is considered, which is appropriate for risk assessment.Flame Partially Bridging the Conductor Gap.If the flame height is lower than the conductor height (Fig. 1(b)), the conductor-to-ground insulation path comprises three segments: the flame region, a smoke-filled region above the flame, and the vegetation (approximated as a conductor). The phase-to-phase insulation path is entirely surrounded by smoke. Thus, the withstand voltages are calculated as the sum of the withstand capabilities of the different segments:18\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U_{\mathrm{g}}}={H_{\mathrm{f}}}{E_{\mathrm{f}}}+({H_{\mathrm{l}}} - {H_{\mathrm{t}}} - {H_{\mathrm{f}}}){E_{\mathrm{s}}}$$\end{document}19\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U_{\mathrm{p}}}={d_{\mathrm{p}}}{E_{\mathrm{s}}}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H_{\mathrm{f}}}$$\end{document} is the computed flame height from the fire model; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E_{\mathrm{s}}}$$\end{document} is the average smoke withstand field strength under wildfire conditions.The current physical model employs area-averaged values for vegetation properties (e.g., flame temperature, particle correction factor) derived from experimental studies, and does not explicitly account for small-scale, spatial heterogeneity in combustion processes. While this simplification may introduce local uncertainty, the use of representative values ensures the model’s reliability for regional-scale risk assessment. Future model iterations will incorporate spatially distributed vegetation types and stochastic combustion modeling to better capture this variability. Flame Fully Bridging the Conductor Gap.If the vegetation flame height exceeds the conductor height (considering sag), as depicted in Fig. 1(a), both the conductor-to-ground and phase-to-phase insulation paths are assumed to be fully immersed in the flame. Given the rapid carbonization of burning wood and its resulting high electrical conductivity compared to air, the flaming vegetation is approximated as a conductor. Consequently, the withstand voltages Ug and Up are determined primarily by the electrical strength of the flame itself.16\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U_{\mathrm{g}}}=({H_{\mathrm{l}}} - {H_{\mathrm{t}}}){E_{\mathrm{f}}}={d_{{\mathrm{lt}}}}{E_{\mathrm{f}}}$$\end{document}17\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U_{\mathrm{p}}}={d_{\mathrm{p}}}{E_{\mathrm{f}}}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H_{\mathrm{l}}}$$\end{document} is the conductor height above ground, accounting for sag; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H_{\mathrm{t}}}$$\end{document}is the maximum canopy height beneath the line; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d_{{\mathrm{lt}}}}$$\end{document} is the length of the air gap between the conductor and the canopy top; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d_{\mathrm{p}}}$$\end{document} is the phase-to-phase distance; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E_{\mathrm{f}}}$$\end{document} is the average withstand field strength of the flame. Note that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d_{{\mathrm{lt}}}}$$\end{document} represents the vertical height in the present study. While flame tilt due to wind is not explicitly modeled here, the conservative assumption of vertical flame height ensures that the worst-case scenario for gap bridging is considered, which is appropriate for risk assessment. Flame Fully Bridging the Conductor Gap. If the vegetation flame height exceeds the conductor height (considering sag), as depicted in Fig. 1(a), both the conductor-to-ground and phase-to-phase insulation paths are assumed to be fully immersed in the flame. Given the rapid carbonization of burning wood and its resulting high electrical conductivity compared to air, the flaming vegetation is approximated as a conductor. Consequently, the withstand voltages Ug and Up are determined primarily by the electrical strength of the flame itself.16\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U_{\mathrm{g}}}=({H_{\mathrm{l}}} - {H_{\mathrm{t}}}){E_{\mathrm{f}}}={d_{{\mathrm{lt}}}}{E_{\mathrm{f}}}$$\end{document}17\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U_{\mathrm{p}}}={d_{\mathrm{p}}}{E_{\mathrm{f}}}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H_{\mathrm{l}}}$$\end{document} is the conductor height above ground, accounting for sag; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H_{\mathrm{t}}}$$\end{document}is the maximum canopy height beneath the line; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d_{{\mathrm{lt}}}}$$\end{document} is the length of the air gap between the conductor and the canopy top; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d_{\mathrm{p}}}$$\end{document} is the phase-to-phase distance; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E_{\mathrm{f}}}$$\end{document} is the average withstand field strength of the flame. Note that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d_{{\mathrm{lt}}}}$$\end{document} represents the vertical height in the present study. While flame tilt due to wind is not explicitly modeled here, the conservative assumption of vertical flame height ensures that the worst-case scenario for gap bridging is considered, which is appropriate for risk assessment. Flame Partially Bridging the Conductor Gap.If the flame height is lower than the conductor height (Fig. 1(b)), the conductor-to-ground insulation path comprises three segments: the flame region, a smoke-filled region above the flame, and the vegetation (approximated as a conductor). The phase-to-phase insulation path is entirely surrounded by smoke. Thus, the withstand voltages are calculated as the sum of the withstand capabilities of the different segments:18\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U_{\mathrm{g}}}={H_{\mathrm{f}}}{E_{\mathrm{f}}}+({H_{\mathrm{l}}} - {H_{\mathrm{t}}} - {H_{\mathrm{f}}}){E_{\mathrm{s}}}$$\end{document}19\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U_{\mathrm{p}}}={d_{\mathrm{p}}}{E_{\mathrm{s}}}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H_{\mathrm{f}}}$$\end{document} is the computed flame height from the fire model; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E_{\mathrm{s}}}$$\end{document} is the average smoke withstand field strength under wildfire conditions.The current physical model employs area-averaged values for vegetation properties (e.g., flame temperature, particle correction factor) derived from experimental studies, and does not explicitly account for small-scale, spatial heterogeneity in combustion processes. While this simplification may introduce local uncertainty, the use of representative values ensures the model’s reliability for regional-scale risk assessment. Future model iterations will incorporate spatially distributed vegetation types and stochastic combustion modeling to better capture this variability. Flame Partially Bridging the Conductor Gap. If the flame height is lower than the conductor height (Fig. 1(b)), the conductor-to-ground insulation path comprises three segments: the flame region, a smoke-filled region above the flame, and the vegetation (approximated as a conductor). The phase-to-phase insulation path is entirely surrounded by smoke. Thus, the withstand voltages are calculated as the sum of the withstand capabilities of the different segments:18\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U_{\mathrm{g}}}={H_{\mathrm{f}}}{E_{\mathrm{f}}}+({H_{\mathrm{l}}} - {H_{\mathrm{t}}} - {H_{\mathrm{f}}}){E_{\mathrm{s}}}$$\end{document}19\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U_{\mathrm{p}}}={d_{\mathrm{p}}}{E_{\mathrm{s}}}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H_{\mathrm{f}}}$$\end{document} is the computed flame height from the fire model; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E_{\mathrm{s}}}$$\end{document} is the average smoke withstand field strength under wildfire conditions. The current physical model employs area-averaged values for vegetation properties (e.g., flame temperature, particle correction factor) derived from experimental studies, and does not explicitly account for small-scale, spatial heterogeneity in combustion processes. While this simplification may introduce local uncertainty, the use of representative values ensures the model’s reliability for regional-scale risk assessment. Future model iterations will incorporate spatially distributed vegetation types and stochastic combustion modeling to better capture this variability. Evaluation of wildfire occurrence probability using bayesian model Factors influencing wildfire occurrence This study identifies and collects data for 16 factors influencing wildfire occurrence within the study area, categorized into four groups: geographical conditions, meteorological conditions, combustible conditions, and human activities, as listed in Table 2. The 16 factors are widely recognized as the primary drivers of wildfire ignition and spread in the literature50–52. While no model can encompass every possible factor, our selection represents a holistic set of predictors that capture the key physical, environmental, and anthropogenic processes influencing wildfire occurrence near transmission corridors. Geographical and meteorological factors represent regional natural environmental variables. Geographical factors include elevation, slope, aspect, and distance to water source. Meteorological factors comprise monthly average rainfall, monthly average temperature, monthly average air humidity, and monthly thunderstorm day count. Table 2Environmental factors influencing wildfire occurrence and data sources.Influencing factorData typeData source1 Geographical conditions 1.1 ElevationContinuousRESDC 1.2 SlopeContinuousRESDC 1.3 AspectContinuousRESDC 1.4 Distance to water sourceContinuousRESDC2 Meteorological conditions 2.1 Monthly avg. rainfallContinuousNMICC 2.2 Monthly avg. temperatureContinuousNMICC 2.3 Monthly avg. humidityContinuousNMICC 2.4 Monthly thunderstorm daysContinuousNMICC3 Combustible conditions 3.1 Land use typeDiscreteRESDC 3.2 Vegetation typeDiscreteRESDC 3.3 Fuel loadContinuousNMICC4 Human activities 4.1 Distance to roadsContinuousRESDC 4.2 Distance to residential areasContinuousRESDC 4.3 Population densityContinuousRESDC 4.4 Regional GDP per capitaContinuousRESDC 4.5 Historical fire densityContinuousNMICC Table 2Environmental factors influencing wildfire occurrence and data sources.Influencing factorData typeData source1 Geographical conditions 1.1 ElevationContinuousRESDC 1.2 SlopeContinuousRESDC 1.3 AspectContinuousRESDC 1.4 Distance to water sourceContinuousRESDC2 Meteorological conditions 2.1 Monthly avg. rainfallContinuousNMICC 2.2 Monthly avg. temperatureContinuousNMICC 2.3 Monthly avg. humidityContinuousNMICC 2.4 Monthly thunderstorm daysContinuousNMICC3 Combustible conditions 3.1 Land use typeDiscreteRESDC 3.2 Vegetation typeDiscreteRESDC 3.3 Fuel loadContinuousNMICC4 Human activities 4.1 Distance to roadsContinuousRESDC 4.2 Distance to residential areasContinuousRESDC 4.3 Population densityContinuousRESDC 4.4 Regional GDP per capitaContinuousRESDC 4.5 Historical fire densityContinuousNMICC Environmental factors influencing wildfire occurrence and data sources. The presence of an ignition source and sufficient combustible material are necessary preconditions for wildfires. Research indicates that over 95% of wildfire ignitions are attributed to human activities53. Therefore, five factors were selected to quantify human influence: distance to roads, distance to residential areas, population density, regional GDP per capita, and historical fire point density. Combustible conditions are characterized by vegetation attributes under the transmission line corridor, specifically land use type, vegetation type, and fuel load. Note that the vegetation data are considered within a 30-meter buffer on either side of the transmission corridor to account for potential fire spread from adjacent areas, consistent with typical right-of-way management practices. Data for these 16 factors were sourced from open-access databases provided by the Resource and Environmental Science and Data Center (RESDC) of the Chinese Academy of Sciences (https://www.resdc.cn/) and the National Meteorological Information Center of China (NMICC) (https://sci.cma.cn/). To facilitate Bayesian probability modeling, continuous variables were discretized using frequency-based methods. The Relief algorithm54 is then employed to rank the importance of each influencing factor. The algorithm operates on a dataset comprising n samples (instances), each described by m features and a class label (fire or non-fire), following four steps below: Initialization: Assign an initial weight (often zero) to each feature f.Iterative Sampling: For T iterations, Randomly select a reference sample R; Find the nearest sample from the same class as R (Near Hit, NH); Find the nearest sample from the opposite classes to R (Near Miss, NM – often one per class, then take the closest overall).Weight Update: For each feature f, decrease the weight proportionally to the difference between R and NH on feature f (penalizing features that separate instances of the same class); Increase the weight proportionally to the difference between R and NM on feature f (rewarding features that distinguish between classes).Result: After T iterations, each feature has a final weight. Features are ranked by these weights in descending order to indicate importance. Initialization: Assign an initial weight (often zero) to each feature f. Initialization: Assign an initial weight (often zero) to each feature f. Iterative Sampling: For T iterations, Randomly select a reference sample R; Find the nearest sample from the same class as R (Near Hit, NH); Find the nearest sample from the opposite classes to R (Near Miss, NM – often one per class, then take the closest overall). Iterative Sampling: For T iterations, Randomly select a reference sample R; Find the nearest sample from the same class as R (Near Hit, NH); Find the nearest sample from the opposite classes to R (Near Miss, NM – often one per class, then take the closest overall). Weight Update: For each feature f, decrease the weight proportionally to the difference between R and NH on feature f (penalizing features that separate instances of the same class); Increase the weight proportionally to the difference between R and NM on feature f (rewarding features that distinguish between classes). Weight Update: For each feature f, decrease the weight proportionally to the difference between R and NH on feature f (penalizing features that separate instances of the same class); Increase the weight proportionally to the difference between R and NM on feature f (rewarding features that distinguish between classes). Result: After T iterations, each feature has a final weight. Features are ranked by these weights in descending order to indicate importance. Result: After T iterations, each feature has a final weight. Features are ranked by these weights in descending order to indicate importance. Wildfire occurrence probability for transmission lines Factors with the lowest importance scores are iteratively removed to identify an optimal subset for constructing a parsimonious yet effective Bayesian network model. This optimized model is then used to assess the spatial distribution of wildfire occurrence risk probability along all transmission line corridors within the target region. The overall procedure is illustrated in Fig. 2 and involves the following detailed steps: Fig. 2Flowchart for evaluating wildfire occurrence probability along transmission lines based on the Bayesian model. Flowchart for evaluating wildfire occurrence probability along transmission lines based on the Bayesian model. Stage 1: Data preparation and processing. Data Collection: Gather data for all 16 influencing factors (Table 2) across the target area.Data Processing and Gridding: Discretize the study area into a regular grid (e.g., 1 km × 1 km cells). Assign historical fire locations to corresponding grid cells (positive samples). Randomly select or identify non-fire grid cells (negative samples). Preprocess all factor data (e.g., discretization of continuous variables, normalization/standardization if needed) to prepare for Bayesian network learning and the Relief algorithm. Data Collection: Gather data for all 16 influencing factors (Table 2) across the target area. Data Collection: Gather data for all 16 influencing factors (Table 2) across the target area. Data Processing and Gridding: Discretize the study area into a regular grid (e.g., 1 km × 1 km cells). Assign historical fire locations to corresponding grid cells (positive samples). Randomly select or identify non-fire grid cells (negative samples). Preprocess all factor data (e.g., discretization of continuous variables, normalization/standardization if needed) to prepare for Bayesian network learning and the Relief algorithm. Data Processing and Gridding: Discretize the study area into a regular grid (e.g., 1 km × 1 km cells). Assign historical fire locations to corresponding grid cells (positive samples). Randomly select or identify non-fire grid cells (negative samples). Preprocess all factor data (e.g., discretization of continuous variables, normalization/standardization if needed) to prepare for Bayesian network learning and the Relief algorithm. Stage 2: Feature selection and model training. 3.Feature Importance Ranking: Apply the Relief algorithm to the processed dataset to compute importance scores and rank all influencing factors.4.Remove Least Important Factor: Discard the factor with the lowest current importance ranking.5.Build and Evaluate Model: Using the remaining feature set, partition the data into training and testing sets. Construct a Bayesian network model using the training data. Evaluate the model’s performance on the test set using appropriate metrics (e.g., accuracy, AUC).6.Iteration Decision: Assess if model performance (e.g., AUC) improved after the feature removal: If Yes: Return to step 4, remove the next least important factor from the current subset, and repeat steps 5 and 6; If No: Conclude that the previous feature subset (before the last removal) yielded peak performance. This subset is considered optimal.7.Determine Optimal Model: Finalize the optimal feature combination and the corresponding Bayesian network model trained on that combination. 3.Feature Importance Ranking: Apply the Relief algorithm to the processed dataset to compute importance scores and rank all influencing factors. Feature Importance Ranking: Apply the Relief algorithm to the processed dataset to compute importance scores and rank all influencing factors. 4.Remove Least Important Factor: Discard the factor with the lowest current importance ranking. Remove Least Important Factor: Discard the factor with the lowest current importance ranking. 5.Build and Evaluate Model: Using the remaining feature set, partition the data into training and testing sets. Construct a Bayesian network model using the training data. Evaluate the model’s performance on the test set using appropriate metrics (e.g., accuracy, AUC). Build and Evaluate Model: Using the remaining feature set, partition the data into training and testing sets. Construct a Bayesian network model using the training data. Evaluate the model’s performance on the test set using appropriate metrics (e.g., accuracy, AUC). 6.Iteration Decision: Assess if model performance (e.g., AUC) improved after the feature removal: If Yes: Return to step 4, remove the next least important factor from the current subset, and repeat steps 5 and 6; If No: Conclude that the previous feature subset (before the last removal) yielded peak performance. This subset is considered optimal. Iteration Decision: Assess if model performance (e.g., AUC) improved after the feature removal: If Yes: Return to step 4, remove the next least important factor from the current subset, and repeat steps 5 and 6; If No: Conclude that the previous feature subset (before the last removal) yielded peak performance. This subset is considered optimal. 7.Determine Optimal Model: Finalize the optimal feature combination and the corresponding Bayesian network model trained on that combination. Determine Optimal Model: Finalize the optimal feature combination and the corresponding Bayesian network model trained on that combination. Stage 3: Application and output. 8.Bayesian Modeling and Application: Utilize the finalized optimal Bayesian network model with the selected features.9.Output Risk Assessment: For each transmission line corridor segment (e.g., between towers), input the relevant values of the optimal influencing factors into the model. The model computes and outputs the wildfire occurrence probability for each segment, thereby generating a spatial risk assessment along the entire line. 8.Bayesian Modeling and Application: Utilize the finalized optimal Bayesian network model with the selected features. Bayesian Modeling and Application: Utilize the finalized optimal Bayesian network model with the selected features. 9.Output Risk Assessment: For each transmission line corridor segment (e.g., between towers), input the relevant values of the optimal influencing factors into the model. The model computes and outputs the wildfire occurrence probability for each segment, thereby generating a spatial risk assessment along the entire line. Output Risk Assessment: For each transmission line corridor segment (e.g., between towers), input the relevant values of the optimal influencing factors into the model. The model computes and outputs the wildfire occurrence probability for each segment, thereby generating a spatial risk assessment along the entire line. This structured approach employs iterative feature selection to eliminate redundancy, enhancing model simplicity and efficiency, and ultimately delivers an accurate and interpretable assessment of wildfire probability. A sensitivity analysis of the Bayesian network is conducted by varying the input probabilities of the top three factors (distance to residential areas, vegetation type, monthly rainfall). The model output is most sensitive to changes in vegetation type and rainfall, reinforcing their importance as key predictors. Susceptibility assessment of wildfire-induced transmission line tripping Tripping risk The occurrence of a wildfire-induced transmission line trip requires the conjunction of two conditions: firstly, a wildfire ignites sufficiently close to the line to degrade the air insulation integrity beneath the conductors; secondly, the reduced withstand strength of either the conductor-to-ground or phase-to-phase air gap falls below the prevailing operating voltage stress, leading to insulation breakdown and consequent tripping. Given the typical duration of wildfires, successful automatic reclosing is often unlikely; therefore, the potential mitigating effect of reclosing is not considered in this risk assessment. Section 3 detailed the development of a Bayesian evaluation model that incorporates geographical, meteorological, fuel, and human activity factors to compute the spatial probability of wildfire occurrence (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${P_{\mathrm{f}}}$$\end{document}) near the line. Building on this, and integrating the line insulation breakdown risk (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R_{{\mathrm{tf}}}}$$\end{document}) during wildfire conditions derived in Sect. 2, the overall wildfire-induced tripping risk (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R_{\mathrm{t}}}$$\end{document}) for a transmission line section is calculated as the product:20\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R_{\mathrm{t}}}={P_{\mathrm{f}}} \times {R_{{\mathrm{tf}}}}$$\end{document} Susceptibility assessment procedure Leveraging the aforementioned tripping risk assessment model, the wildfire tripping risk can be quantified for all sections within the target study area. Based on the calculated risk values, sections are classified into distinct risk levels, and a spatial risk distribution map is generated. The procedure, outlined in Fig. 3, consists of the following steps: Data Collection and Preprocessing: Wildfire alert levels are highest when fires occur within 1 km of a line. Consequently, the study area is partitioned into a grid (e.g., 1 km × 1 km cells). Data for the influencing factors listed in Table 2 are acquired from the RESDC and NMICC. Factors with coarser spatial resolution than the grid size are interpolated (e.g., using Kriging) to the grid level. Continuous variables are discretized as required.Optimal Bayesian Wildfire Probability Model Construction: Grid cells with historical wildfire occurrences serve as positive samples. The wildfire dataset includes both natural (e.g., lightning) and human-induced ignitions. An equivalent number of cells without historical fires are randomly selected as negative samples. The Relief algorithm ranks the importance of the wildfire influencing factors. Factors are iteratively removed based on this ranking (starting with the least important), and the Bayesian model is retrained and validated each time, to identify the optimal predictive model with the most relevant feature set.Wildfire Probability Calculation for Transmission Corridors: Gather transmission line ledger data (geographic coordinates, voltage level, etc.) from the utility company. Employ the optimal Bayesian model to compute the wildfire occurrence probability (Pf) for every grid cell in the study area. For each transmission line segment (e.g., defined between two adjacent towers), determine the representative wildfire probability, often taken at the corridor midpoint or averaged over relevant grid cells.Insulation Breakdown Risk Calculation under Wildfire Conditions: Utilize LiDAR point cloud technology or other sources to obtain detailed transmission corridor data, including conductor height above ground (considering sag), maximum vegetation (canopy) height, and phase spacing for each line section. Estimate the maximum potential flame height (Hf) for vegetation under the corridor using the fire model. Determine if flames would fully or partially bridge the conductor gap based on Hf, conductor height, and vegetation height. Calculate the appropriate withstand voltages (Ug, Up) using the formulas from Sect. 2.3 and the corresponding withstand field strengths (Ef, Es). Compute the insulation breakdown risk Rtf using the relevant formulas from Sect. 2.3.Susceptibility Assessment and Risk Mapping: Calculate the tripping risk Rt for each line section using Eq. (20). Normalize the Rt values if necessary. Classify the sections into risk levels (e.g., I-IV) using an equal-interval method or other suitable classification scheme (see Table 3). Employ GIS software (e.g., ArcGIS) to create a visual map showing the spatial distribution of tripping risk levels along the transmission lines. Data Collection and Preprocessing: Wildfire alert levels are highest when fires occur within 1 km of a line. Consequently, the study area is partitioned into a grid (e.g., 1 km × 1 km cells). Data for the influencing factors listed in Table 2 are acquired from the RESDC and NMICC. Factors with coarser spatial resolution than the grid size are interpolated (e.g., using Kriging) to the grid level. Continuous variables are discretized as required. Data Collection and Preprocessing: Wildfire alert levels are highest when fires occur within 1 km of a line. Consequently, the study area is partitioned into a grid (e.g., 1 km × 1 km cells). Data for the influencing factors listed in Table 2 are acquired from the RESDC and NMICC. Factors with coarser spatial resolution than the grid size are interpolated (e.g., using Kriging) to the grid level. Continuous variables are discretized as required. Optimal Bayesian Wildfire Probability Model Construction: Grid cells with historical wildfire occurrences serve as positive samples. The wildfire dataset includes both natural (e.g., lightning) and human-induced ignitions. An equivalent number of cells without historical fires are randomly selected as negative samples. The Relief algorithm ranks the importance of the wildfire influencing factors. Factors are iteratively removed based on this ranking (starting with the least important), and the Bayesian model is retrained and validated each time, to identify the optimal predictive model with the most relevant feature set. Optimal Bayesian Wildfire Probability Model Construction: Grid cells with historical wildfire occurrences serve as positive samples. The wildfire dataset includes both natural (e.g., lightning) and human-induced ignitions. An equivalent number of cells without historical fires are randomly selected as negative samples. The Relief algorithm ranks the importance of the wildfire influencing factors. Factors are iteratively removed based on this ranking (starting with the least important), and the Bayesian model is retrained and validated each time, to identify the optimal predictive model with the most relevant feature set. Wildfire Probability Calculation for Transmission Corridors: Gather transmission line ledger data (geographic coordinates, voltage level, etc.) from the utility company. Employ the optimal Bayesian model to compute the wildfire occurrence probability (Pf) for every grid cell in the study area. For each transmission line segment (e.g., defined between two adjacent towers), determine the representative wildfire probability, often taken at the corridor midpoint or averaged over relevant grid cells. Wildfire Probability Calculation for Transmission Corridors: Gather transmission line ledger data (geographic coordinates, voltage level, etc.) from the utility company. Employ the optimal Bayesian model to compute the wildfire occurrence probability (Pf) for every grid cell in the study area. For each transmission line segment (e.g., defined between two adjacent towers), determine the representative wildfire probability, often taken at the corridor midpoint or averaged over relevant grid cells. Insulation Breakdown Risk Calculation under Wildfire Conditions: Utilize LiDAR point cloud technology or other sources to obtain detailed transmission corridor data, including conductor height above ground (considering sag), maximum vegetation (canopy) height, and phase spacing for each line section. Estimate the maximum potential flame height (Hf) for vegetation under the corridor using the fire model. Determine if flames would fully or partially bridge the conductor gap based on Hf, conductor height, and vegetation height. Calculate the appropriate withstand voltages (Ug, Up) using the formulas from Sect. 2.3 and the corresponding withstand field strengths (Ef, Es). Compute the insulation breakdown risk Rtf using the relevant formulas from Sect. 2.3. Insulation Breakdown Risk Calculation under Wildfire Conditions: Utilize LiDAR point cloud technology or other sources to obtain detailed transmission corridor data, including conductor height above ground (considering sag), maximum vegetation (canopy) height, and phase spacing for each line section. Estimate the maximum potential flame height (Hf) for vegetation under the corridor using the fire model. Determine if flames would fully or partially bridge the conductor gap based on Hf, conductor height, and vegetation height. Calculate the appropriate withstand voltages (Ug, Up) using the formulas from Sect. 2.3 and the corresponding withstand field strengths (Ef, Es). Compute the insulation breakdown risk Rtf using the relevant formulas from Sect. 2.3. Susceptibility Assessment and Risk Mapping: Calculate the tripping risk Rt for each line section using Eq. (20). Normalize the Rt values if necessary. Classify the sections into risk levels (e.g., I-IV) using an equal-interval method or other suitable classification scheme (see Table 3). Employ GIS software (e.g., ArcGIS) to create a visual map showing the spatial distribution of tripping risk levels along the transmission lines. Susceptibility Assessment and Risk Mapping: Calculate the tripping risk Rt for each line section using Eq. (20). Normalize the Rt values if necessary. Classify the sections into risk levels (e.g., I-IV) using an equal-interval method or other suitable classification scheme (see Table 3). Employ GIS software (e.g., ArcGIS) to create a visual map showing the spatial distribution of tripping risk levels along the transmission lines. Fig. 3Flowchart for the susceptibility assessment of wildfire-induced transmission line tripping risk. Flowchart for the susceptibility assessment of wildfire-induced transmission line tripping risk. Table 3Classification scheme for wildfire-induced transmission line tripping risk levels.Tripping risk (Rt)[0, 0.25)[0.25, 0.5)[0.5, 0.75)[0.75, 1]Risk levelLevel ILevel IILevel IIILevel IVInterpretationLowModerateHighVery high Table 3Classification scheme for wildfire-induced transmission line tripping risk levels.Tripping risk (Rt)[0, 0.25)[0.25, 0.5)[0.5, 0.75)[0.75, 1]Risk levelLevel ILevel IILevel IIILevel IVInterpretationLowModerateHighVery high Classification scheme for wildfire-induced transmission line tripping risk levels. Case study and results Study Area.This study focused on the power grid of Guangdong Province, China. The provincial distribution of wildfire occurrence probability and the subsequent tripping risk for transmission line sections were evaluated. Guangdong features a subtropical monsoon climate with hot summers, mild winters, distinct seasons, and abundant vegetation. Notably, 57.9% of the province’s terrain is mountainous or hilly. This complex topography can hinder satellite-based fire detection due to occlusion, increasing the challenge of maintaining grid stability. Fig. 4Importance ranking of wildfire occurrence influencing factors based on the relief algorithm. Factor Importance Ranking.Data for the 16 predefined factors were compiled, and their relative importance was evaluated using the Relief algorithm. The results, presented in Fig. 4, identify distance to residential areas, vegetation type, and monthly average rainfall as the three most significant factors affecting wildfire occurrence, with substantially higher importance scores than the others. This suggests: (i) a strong link between human activity (proximity to residences) and ignition sources; (ii) the critical role of fuel type (vegetation) in ignition and spread dynamics; and (iii) the importance of moisture availability (rainfall) in inhibiting fire occurrence both directly and indirectly through vegetation growth. Transmission line maintenance strategies should prioritize monitoring these key factors.Wildfire Occurrence Probability Assessment.Historical wildfire incidents within the province from 2012 to 2022 were gathered. Data from 2012 to 2016 were used to compute the historical fire density feature for each grid cell. Incidents from 2017 to 2021 served as positive samples for model training and testing. Data from 2022 were reserved for validation. Applying the optimal Bayesian model, the wildfire occurrence probability was calculated for each 1 km × 1 km grid cell. The province was categorized into four zones (A, B, C, D) based on increasing probability thresholds (Fig. 5). Historical fire points from the validation period (2022) are overlaid on the map, showing a strong concentration within the high-probability Zone D, which corroborates the model’s predictive validity. Fig. 5Wildfire occurrence risk probability distribution across Guangdong Province, China. The spatial distribution of wildfire probability exhibits several characteristics: Zones A, B, C, and D cover 15.5%, 34.4%, 21.7%, and 28.4% of the province’s area, respectively, indicating significant extents of moderate to high probability. Wildfire probability generally increases from south to north and east to west, with particular concern warranted for the northern mountainous regions exhibiting large contiguous areas of high probability (Zone D). Zone D also displays greater spatial dispersion compared to other zones, highlighting the inherent spatial uncertainty in wildfire ignition locations, which necessitates vigilant monitoring. The overprediction of wildfire probability in northern and eastern Guangdong may be attributed to unmodeled factors such as localized fire suppression efforts, variations in fuel load not captured by vegetation type, or anthropogenic firebreaks. These areas may also have lower ignition frequency despite favorable conditions due to lower human presence.Susceptibility Assessment of Transmission Line Tripping.Integrating the provincial wildfire probability map with transmission line data (location, voltage) and corridor parameters (heights, spacing) from the Guangdong Power Grid, the insulation breakdown risk (Rtf) during wildfires was computed for all sections of 110 kV and above lines. Subsequently, the tripping risk (Rt) was calculated, and sections were classified into susceptibility levels. The proportion of line sections falling into each risk level within zones A-D and overall is depicted in Fig. 6. A strong positive correlation is observed between wildfire probability (increasing from zone A to D) and the proportion of line sections classified as high susceptibility (Levels III & IV), while the proportion of low susceptibility sections (Levels I & II) decreases accordingly. Across the entire province, the proportions of 110 kV and above transmission line sections categorized into Levels I, II, III, and IV for wildfire-induced tripping susceptibility are 0.433, 0.262, 0.190, and 0.115, respectively. This indicates that high-susceptibility sections constitute a smaller fraction compared to low and moderate susceptibility sections. Fig. 6Distribution of tripping risk levels for 110 kV and above transmission lines in Guangdong Province. Individual Case Validation.A specific 220 kV line was examined in detail (Fig. 7). The line runs northeast to southwest. Model assessment assigned 29.5%, 30.2%, 18.9%, and 21.4% of its length to susceptibility Levels I, II, III, and IV, respectively. Fig. 7Susceptibility distribution for wildfire-induced tripping along a section of a 220 kV transmission line in Guangdong Province. Study Area.This study focused on the power grid of Guangdong Province, China. The provincial distribution of wildfire occurrence probability and the subsequent tripping risk for transmission line sections were evaluated. Guangdong features a subtropical monsoon climate with hot summers, mild winters, distinct seasons, and abundant vegetation. Notably, 57.9% of the province’s terrain is mountainous or hilly. This complex topography can hinder satellite-based fire detection due to occlusion, increasing the challenge of maintaining grid stability. Fig. 4Importance ranking of wildfire occurrence influencing factors based on the relief algorithm. Study Area. This study focused on the power grid of Guangdong Province, China. The provincial distribution of wildfire occurrence probability and the subsequent tripping risk for transmission line sections were evaluated. Guangdong features a subtropical monsoon climate with hot summers, mild winters, distinct seasons, and abundant vegetation. Notably, 57.9% of the province’s terrain is mountainous or hilly. This complex topography can hinder satellite-based fire detection due to occlusion, increasing the challenge of maintaining grid stability. Fig. 4Importance ranking of wildfire occurrence influencing factors based on the relief algorithm. Importance ranking of wildfire occurrence influencing factors based on the relief algorithm. Factor Importance Ranking.Data for the 16 predefined factors were compiled, and their relative importance was evaluated using the Relief algorithm. The results, presented in Fig. 4, identify distance to residential areas, vegetation type, and monthly average rainfall as the three most significant factors affecting wildfire occurrence, with substantially higher importance scores than the others. This suggests: (i) a strong link between human activity (proximity to residences) and ignition sources; (ii) the critical role of fuel type (vegetation) in ignition and spread dynamics; and (iii) the importance of moisture availability (rainfall) in inhibiting fire occurrence both directly and indirectly through vegetation growth. Transmission line maintenance strategies should prioritize monitoring these key factors. Factor Importance Ranking. Data for the 16 predefined factors were compiled, and their relative importance was evaluated using the Relief algorithm. The results, presented in Fig. 4, identify distance to residential areas, vegetation type, and monthly average rainfall as the three most significant factors affecting wildfire occurrence, with substantially higher importance scores than the others. This suggests: (i) a strong link between human activity (proximity to residences) and ignition sources; (ii) the critical role of fuel type (vegetation) in ignition and spread dynamics; and (iii) the importance of moisture availability (rainfall) in inhibiting fire occurrence both directly and indirectly through vegetation growth. Transmission line maintenance strategies should prioritize monitoring these key factors. Wildfire Occurrence Probability Assessment.Historical wildfire incidents within the province from 2012 to 2022 were gathered. Data from 2012 to 2016 were used to compute the historical fire density feature for each grid cell. Incidents from 2017 to 2021 served as positive samples for model training and testing. Data from 2022 were reserved for validation. Applying the optimal Bayesian model, the wildfire occurrence probability was calculated for each 1 km × 1 km grid cell. The province was categorized into four zones (A, B, C, D) based on increasing probability thresholds (Fig. 5). Historical fire points from the validation period (2022) are overlaid on the map, showing a strong concentration within the high-probability Zone D, which corroborates the model’s predictive validity. Fig. 5Wildfire occurrence risk probability distribution across Guangdong Province, China. The spatial distribution of wildfire probability exhibits several characteristics: Zones A, B, C, and D cover 15.5%, 34.4%, 21.7%, and 28.4% of the province’s area, respectively, indicating significant extents of moderate to high probability. Wildfire probability generally increases from south to north and east to west, with particular concern warranted for the northern mountainous regions exhibiting large contiguous areas of high probability (Zone D). Zone D also displays greater spatial dispersion compared to other zones, highlighting the inherent spatial uncertainty in wildfire ignition locations, which necessitates vigilant monitoring. The overprediction of wildfire probability in northern and eastern Guangdong may be attributed to unmodeled factors such as localized fire suppression efforts, variations in fuel load not captured by vegetation type, or anthropogenic firebreaks. These areas may also have lower ignition frequency despite favorable conditions due to lower human presence. Wildfire Occurrence Probability Assessment. Historical wildfire incidents within the province from 2012 to 2022 were gathered. Data from 2012 to 2016 were used to compute the historical fire density feature for each grid cell. Incidents from 2017 to 2021 served as positive samples for model training and testing. Data from 2022 were reserved for validation. Applying the optimal Bayesian model, the wildfire occurrence probability was calculated for each 1 km × 1 km grid cell. The province was categorized into four zones (A, B, C, D) based on increasing probability thresholds (Fig. 5). Historical fire points from the validation period (2022) are overlaid on the map, showing a strong concentration within the high-probability Zone D, which corroborates the model’s predictive validity. Fig. 5Wildfire occurrence risk probability distribution across Guangdong Province, China. Wildfire occurrence risk probability distribution across Guangdong Province, China. The spatial distribution of wildfire probability exhibits several characteristics: Zones A, B, C, and D cover 15.5%, 34.4%, 21.7%, and 28.4% of the province’s area, respectively, indicating significant extents of moderate to high probability. Wildfire probability generally increases from south to north and east to west, with particular concern warranted for the northern mountainous regions exhibiting large contiguous areas of high probability (Zone D). Zone D also displays greater spatial dispersion compared to other zones, highlighting the inherent spatial uncertainty in wildfire ignition locations, which necessitates vigilant monitoring. The overprediction of wildfire probability in northern and eastern Guangdong may be attributed to unmodeled factors such as localized fire suppression efforts, variations in fuel load not captured by vegetation type, or anthropogenic firebreaks. These areas may also have lower ignition frequency despite favorable conditions due to lower human presence. Susceptibility Assessment of Transmission Line Tripping.Integrating the provincial wildfire probability map with transmission line data (location, voltage) and corridor parameters (heights, spacing) from the Guangdong Power Grid, the insulation breakdown risk (Rtf) during wildfires was computed for all sections of 110 kV and above lines. Subsequently, the tripping risk (Rt) was calculated, and sections were classified into susceptibility levels. The proportion of line sections falling into each risk level within zones A-D and overall is depicted in Fig. 6. A strong positive correlation is observed between wildfire probability (increasing from zone A to D) and the proportion of line sections classified as high susceptibility (Levels III & IV), while the proportion of low susceptibility sections (Levels I & II) decreases accordingly. Across the entire province, the proportions of 110 kV and above transmission line sections categorized into Levels I, II, III, and IV for wildfire-induced tripping susceptibility are 0.433, 0.262, 0.190, and 0.115, respectively. This indicates that high-susceptibility sections constitute a smaller fraction compared to low and moderate susceptibility sections. Fig. 6Distribution of tripping risk levels for 110 kV and above transmission lines in Guangdong Province. Susceptibility Assessment of Transmission Line Tripping. Integrating the provincial wildfire probability map with transmission line data (location, voltage) and corridor parameters (heights, spacing) from the Guangdong Power Grid, the insulation breakdown risk (Rtf) during wildfires was computed for all sections of 110 kV and above lines. Subsequently, the tripping risk (Rt) was calculated, and sections were classified into susceptibility levels. The proportion of line sections falling into each risk level within zones A-D and overall is depicted in Fig. 6. A strong positive correlation is observed between wildfire probability (increasing from zone A to D) and the proportion of line sections classified as high susceptibility (Levels III & IV), while the proportion of low susceptibility sections (Levels I & II) decreases accordingly. Across the entire province, the proportions of 110 kV and above transmission line sections categorized into Levels I, II, III, and IV for wildfire-induced tripping susceptibility are 0.433, 0.262, 0.190, and 0.115, respectively. This indicates that high-susceptibility sections constitute a smaller fraction compared to low and moderate susceptibility sections. Fig. 6Distribution of tripping risk levels for 110 kV and above transmission lines in Guangdong Province. Distribution of tripping risk levels for 110 kV and above transmission lines in Guangdong Province. Individual Case Validation.A specific 220 kV line was examined in detail (Fig. 7). The line runs northeast to southwest. Model assessment assigned 29.5%, 30.2%, 18.9%, and 21.4% of its length to susceptibility Levels I, II, III, and IV, respectively. Fig. 7Susceptibility distribution for wildfire-induced tripping along a section of a 220 kV transmission line in Guangdong Province. Individual Case Validation. A specific 220 kV line was examined in detail (Fig. 7). The line runs northeast to southwest. Model assessment assigned 29.5%, 30.2%, 18.9%, and 21.4% of its length to susceptibility Levels I, II, III, and IV, respectively. Fig. 7Susceptibility distribution for wildfire-induced tripping along a section of a 220 kV transmission line in Guangdong Province. Susceptibility distribution for wildfire-induced tripping along a section of a 220 kV transmission line in Guangdong Province. During the high-risk spring season of 2022, two wildfires occurred within the line’s corridor (marked on Fig. 7). The first fire (Feb. 16, 11:30) ignited in a location classified as being at the boundary between susceptibility Level III and Level IV. It burned until 13:00 without causing a line trip. The second fire (Mar. 24, 16:00) occurred squarely in a Level IV section. By 17:30, the fire led to a line trip, and reclosing attempts failed, causing significant grid disruption. The vegetation under the line in this section was primarily tall broadleaf forest (~ 18 m height), with a clearance of only ~ 5 m to the conductor. The normalized tripping risk probability reached 82.4%. This real event demonstrates the model’s accuracy and practical utility. Proactive measures informed by such risk maps, like targeted vegetation management within high-risk corridors, could effectively reduce the likelihood of wildfire-related trips. Conclusion This study addressed the challenge of assessing the risk of wildfire-induced transmission line tripping by developing an integrated methodology combining susceptibility zoning and risk quantification. The coupling of a Bayesian wildfire occurrence probability model with a physics-based insulation breakdown risk model enabled a spatially explicit evaluation of tripping risk. The key findings are: Feature importance analysis using the Relief algorithm identified distance to residential areas, vegetation type, and monthly average rainfall as the predominant factors influencing wildfire occurrence within the study area, emphasizing the critical roles of human activity and fuel characteristics. The optimized Bayesian network model effectively predicted spatial patterns of wildfire probability, validated by the strong agreement between high-probability zones and independent historical fire incident data. A scenario-based approach for assessing insulation breakdown risk, incorporating flame height and modified smoke withstand strength, successfully captured the significant impact of flame bridging on the electrical withstand capability of air gaps. The case study in Guangdong Province demonstrated that the resulting tripping risk distribution was logically consistent with geographical features, vegetation types, and crucially, aligned with actual recorded tripping events, confirming the model’s accuracy and operational relevance. The findings provide a scientifically grounded basis for implementing zonal and prioritized wildfire prevention strategies for power grids, thereby enhancing their resilience. It should be noted that the current physical model employs area-averaged parameters for vegetation combustion, which is a simplification suitable for regional-scale assessment but may not capture hyper-local variability. Future work could enhance the model’s temporal dynamics by integrating real-time meteorological data and time-varying vegetation parameters. Furthermore, coupling the assessment with advanced monitoring technologies, such as drones and satellite remote sensing, could facilitate dynamic risk monitoring and early warning systems. While the framework is developed for a subtropical region, it is transferable to other regions provided local data on vegetation, climate, and grid infrastructure are available. Future work should validate the model in diverse biogeographic and climatic settings.