Probability-based modeling for a quantitative wildfire risk analysis in the protected areas of Guilan Province

Extended Abstract

Introduction

Wildfires reveal evidence of forest soil, water, and vegetation disturbances resulting from various interacting natural and human factors that create patterns that vary spatially and temporally. Fire risk assessment allows for identifying these factors and estimating their area of influence, thereby determining locations at high fire risk. Fire risk assessment typically involves the ignition probability (IP) and burn probability (BP) modeling of natural and human-made resources, as well as identifying resource responses to fires of varying severity. In recent decades, wildfires have caused significant damage to the Hyrcanian forests in northern Iran, even in the protected areas. Therefore, this study focuses on the spatial distributions of fire size, fire frequency, IP, and BP as essential components of the fire risk framework. First, a historical fire database (including ignition points, burned area, etc.) was prepared using available resources and field surveying. Second, a modeling approach using a limited number of auxiliary variables representing the fire environment (fuel, topography, and weather) and the historical fires (1992-2022) was implemented to calculate IP and BP. The spatial distribution of these parameters helps improve decision-making in fire prevention and control strategies.

Materials and Methods

Guilan Province is located in northern Iran and has an area of 14,044 km2 with an average elevation of 741 m above sea level. This province has 20 natural protected areas, which cover a total of 256,488 ha. ArcGIS 10.8 was used to create a historical fire database in the study area by digitizing fires between 1992 and 2022 from maps or by importing information directly from the previous GIS datasets. Point process models (PPMs) were used to analyze the spatial distribution of fire frequency. PPMs are a regression approach to model point data (i.e., geographic coordinates) for the number of times a 100-m pixel burns between 1992 and 2022. An existing raster map of the study area was converted to points by calculating the center of each pixel, and each point was assigned a frequency. Furthermore, IP was calculated as the average ignition probability occurring over a year in a raster pixel. To help use the fire ignition density to plan preventive activities, the output values were classified into five classes reflecting ignition occurrences (from very low to very high). Finally, the fire risk using BP was assessed by considering topography, fuel loads, and weather using FlamMap. To calculate BP, 1000 random ignition points were created based on the distribution of historical ignition points in the study area. The maximum fire simulation time was set to 6 hr (the average fire duration in the area).

Results and Discussion

There are 186 recorded fires (total burned area of 2232 ha) with an average annual number of 6 fires (average burned area of 12 ha) in the protected areas. Fires <10 ha accounted for 62.4% and 30.8% of the fire number and the burned area, respectively. Fires (10-50 ha) accounted for 32.8% and 46.2% of the fire number and the burned area, respectively. Fires (50-100 ha) accounted for 2.7% and 14.5% of the fire number and the burned area, respectively. Finally, fires >100 ha accounted for <0.5% of the fire number but alone accounted for 8.4% of the burned area. The distribution of fire frequency ranged from 0 to 6. The largest protected areas (40%) experienced no fires. 13% of these areas had 1-2 fire frequencies. Furthermore, 48% of this area had more than two fire frequencies. About 35% of the study area had very low and low IP values, 36% had medium IP, and 18 and 11% had high and very high IP, respectively. 88% of the study area had low and moderate BP values, and 12% had high and very high predicted values. Two fire regimes can be distinguished in the area, one with relatively high fire frequency and BP (mainly at higher elevations) and the other with low fire frequency and BP (at lower elevations). High fire frequency and BP is very limited in extent and occurs in the patches in the southern area. In contrast, low fire frequency and BP regime is the most widespread regime in the area (except for the southern part).

Conclusion

According to the simulated patterns of fire frequency, IP, and BP in the study area, a clear distinction between the actual historical fire perimeters and the predicted burn pattern is that there are areas of moderate to high IP where fires have not occurred in the past 30 years. This is particularly evident in the southern and central parts of the area, where fires have either not occurred or have been very limited in extent. Therefore, a justifiable assessment could be that the likelihood of fire spread and vegetation communities undergoing extensive and long-term changes following the fire is high shortly. Although this study focuses on the protected forest areas, this approach can be applied to fire risk modeling at larger scales. This allows for broader application in natural resources management and planning at regional and national levels. It also provides a comprehensive tool for assessing and managing forest vegetation, soil, and water vulnerability.