A Spatial Assessment of Wildfire Risk for Transmission-Line Corridor Based on a Weighted Naïve Bayes Model

Kunxuan Xiang, You Zhou, Enze Zhou, Junhan Lu, Hui Liu, Yu HuangFrontiers in Energy Research 2022 · DOI

TL;DR

Researchers have developed a Weighted Naïve Bayes (WNB) model to assess wildfire risks along high-voltage transmission corridors in Southern China. By objectively weighting 12 key environmental and human factors, the model accurately predicted over 76% of 2020’s fire events within high-risk zones, offering a computationally efficient alternative to complex Bayesian networks.

What Problem Does This Paper Address?

As China expands its "power transmission from west to east" infrastructure, high-voltage lines must traverse vast, fire-prone wilderness. When a wildfire erupts near these lines, the resulting heat and soot can ionize the air, causing "flashovers" (electrical breakdowns) that trigger grid failures or cascading blackouts.

Existing risk assessment methods often fall into two traps: they are either too subjective (relying on expert questionnaires) or too computationally heavy (like deep neural networks or complex Bayesian Networks). Standard Naïve Bayes (NB) models are fast but flawed because they assume all risk factors—like humidity and vegetation—are independent of one another. This paper seeks a "Goldilocks" solution: a model that is mathematically simple and fast, yet sophisticated enough to account for the varying importance of different environmental drivers.

Methods

The authors proposed a multi-step pipeline to refine wildfire prediction:

  1. Factor Selection & Ranking: They initially collected 14 factors across four categories: Man-made (e.g., GDP, population density), Surface (e.g., vegetation type, fuel load), Climate (temperature, precipitation), and Topographic (elevation, slope).
  2. Information Gain Ratio (IGR): Instead of guessing which factors matter most, they used IGR to rank them. They discovered that "Vegetation Type" and "Land-usage" were the most critical, while "Aspect" (the direction a slope faces) and "Distance from Roads" contributed the least.
  3. The Weighted Naïve Bayes (WNB) Model: To improve upon standard Bayes, they used the Entropy Method to assign objective weights to each factor. This "weakens" the unrealistic independence assumption of standard NB by emphasizing the most influential variables.
  4. Optimization: By iteratively deleting the lowest-ranked factors, they identified an Optimal Factor Set (OFS) of 12 variables that maximized predictive accuracy.

Datasets

The study focused on four southern Chinese provinces: Yunnan, Guizhou, Guangxi, and Guangdong.

  • Fire-Spot Data: Historical data from 2010–2019 (provided by the National Meteorological Center) was used for training, while 825 monitored fire spots from 2020 served as the validation set.
  • Environmental Data: Sourced from the Resource and Environmental Science and Data Center, including Digital Elevation Models (DEM), Normalized Difference Vegetation Index (NDVI), and socio-economic data (GDP/Population).
  • Spatial Resolution: The entire study area was meshed into a 1 km × 1 km grid.

Key Insights and Ideas

  • Pruning for Performance: One of the most interesting findings was that "more data" isn't always better. Removing the two least important factors (Aspect and Distance from Roads) actually increased the model's accuracy and recall. This suggests that low-impact variables act as "noise" in Bayesian calculations.
  • The Power of Vegetation: Vegetation type, land-usage, and historical fire-spot density were identified as the three most significant drivers of risk. This highlights that what is on the ground (fuel) is often more predictive of a fire's start than the specific shape of the terrain (slope/aspect).
  • Efficiency vs. Accuracy: While a full Bayesian Network (BNW) is slightly more accurate because it maps complex inter-dependencies between factors, the WNB model achieved nearly identical performance with a much simpler structure and faster calculation speed.

Key Results

  • Predictive Accuracy: The final WNB model achieved an accuracy ($P_a$) of 0.7566 and a precision ($P_p$) of 0.7728.
  • Validation: When tested against actual fires in 2020, 76.36% of fire events fell into the "high-risk" or "very-high-risk" zones identified by the model.
  • Model Comparison: The Area Under the Curve (AUC) for the models were:
  • BNW (Complex): 0.8446
    • WNB (Proposed): 0.8383
    • NB (Standard): 0.7973
  • Efficiency: The WNB model exhibited the highest "predicting efficiency" (a ratio of accuracy to the area of high-risk zones), making it a highly practical tool for utility companies to prioritize line inspections.

Limitations and Future Work

  • Static vs. Dynamic Data: While the model uses annual climate averages, wildfires are highly dependent on real-time weather shifts (e.g., a sudden dry spell). The authors note that integrating real-time satellite monitoring is a separate prevention measure, but the model itself remains largely static.
  • Human Factors: While GDP and population density are used as proxies for human activity, more granular data on seasonal agricultural burning or specific land-management practices could further refine the "Man-made Factors" category.
  • Future Directions: The authors suggest this framework can be integrated into power grid management systems to improve the implementation efficiency of fire-fighting measures in high-risk corridors.

Summary updated

Abstract

In order to improve the wildfire prevention capacity of transmission lines, a wildfire risk assessment method for transmission-line corridors based on Weighted Naïve Bayes (WNB) is proposed in this paper. Firstly, the importance of 14 collected types of wildfire-related factors is analyzed based on the information gain ratio. Then, the optimal factors set and the most accurate sampling table are constructed by deleting the factors in the lowest order of importance one by one. Finally, the performance of the WNB model is compared to that of NB and BNW models according to the ROC curve and visualization. A total of 76.36% of fire events in 2020 fell in high-risk and very-high-risk regions, indicating the acceptable accuracy of the proposed assessment method of wildfire risk.
BibTeX
@inproceedings{xiang2022spatial,
  author = {Kunxuan Xiang and You Zhou and Enze Zhou and Junhan Lu and Hui Liu and Yu Huang},
  title = {{A Spatial Assessment of Wildfire Risk for Transmission-Line Corridor Based on a Weighted Naïve Bayes Model}},
  booktitle = {Frontiers in Energy Research},
  year = {2022},
  doi = {10.3389/fenrg.2022.829934},
}
Figure