مدل‌سازی ریسک آتش‌سوزی با استفاده از روش‌های سنجش‌ازدور و شبیه‌سازی رفتار آتش‌ در استان گیلان | Wildfire risk modeling using remote sensing methods and fire behavior simulation in Guilan province

اللغة الفارسية. استان گیلان یکی از شاخص­ترین نمونه­های ریسک آتش­سوزی جنگل­ها و مراتع در ایران است. در این تحقیق، شاخص ریسک آتش‌سوزی در این استان بر اساس داده‌های مکانی و تصاویر ماهواره‌ای سنجش‌ازدور که عوامل انسانی و بیوفیزیکی سیمای سرزمین را توصیف می‌کند، تهیه شد. با استفاده از تصاویر ماهواره­ای Landsat-8 OLI/TIRS L1TP سال 2021 و نیز الگوریتم طبقه‌بندی ماشین بردار پشتیبان (SVM)، طبقه‌بندی نظارت شده این تصاویر برای تهیه نقشه پوشش زمین منطقه مورد مطالعه انجام شد. شاخص ریسک آتش‌سوزی (دامنه ارزش از 0/0 تا 1) بر اساس 1) رفتار آتش­سوزی (شدت خط آتش و نرخ گسترش آتش که توسط مدل شبیه­ساز آتش FlamMap (Finney, 2006) مشخص می­شود)؛ 2) تاریخچه آتش­سوزی (مربوط به داده­های آماری آتش­سوزی طی سال­های 1992 تا 2022)؛ و 3) عوامل انسانی (با استفاده از رابطه معکوس این تأثیر با فاصله اقلیدسی از جاده­ها و سکونت­گاه­ها) محاسبه شد. با توجه به نتایج مدل‌سازی، ریسک آتش در منطقه مورد مطالعه از 0/0 تا 68/0 با میانگین 11/0 متغیر بود. مقدار کمینه شاخص ریسک آتش در بخش‌های پست و کم ارتفاع که عمدتاً توسط مدل­های ماده سوختنی غیرقابل سوختن (اراضی لخت، باغ-زراعت آبی، مناطق انسان­ساز و مناطق آبی) پوشیده شده است، مشاهده می­شود. درحالی‌که مقدار بیشینه شاخص ریسک آتش در ارتفاعات، جایی که پوشش­های جنگلی متراکم و با دسترسی محدود به جاده و تجهیزات اطفای آتش در مقایسه با مناطق کم ارتفاع وجود دارد، مشاهده می­شود. شاخص ریسک آتش‌سوزی محاسبه‌شده نشان داد که حدود 21 درصد از استان، ازجمله مهم‌ترین بوم­سازگان­های جنگلی اولیه و کهن­رُست از نظر اکولوژیکی، در برابر آتش‌سوزی‌های جنگلی آسیب‌پذیر است. بر اساس این نتایج، در این مناطق باید منابع مدیریت آتش با تمرکز بیشتری کنترل و برنامه­ریزی شود، زیرساخت­های پیشگیری از آتش تقویت شود، آموزش پیشگیری از آتش برای ارتقای آگاهی مردم در برابر آتش انجام شود و اقدامات مختلفی برای افزایش تاب­آوری بوم­سازگان­های آسیب­پذیر این مناطق باید در نظر گرفته شود.

إنجليزي. Extended Abstract IntroductionIn Iran, various fire spread models have been used to predict fire behavior and spread at regional and local scales (Alhaj Khalaf, Shataee Joibary, Jahdi & Bacciu, 2021; Jahdi, Salis, Alcasena & Del Giudice, 2023). However, the calculation of an integrated fire risk index using these fire spread models has not yet been conducted. Typically, fire risk assessments are based on a set of fire spread descriptors that are used to map fire-prone areas and identify the most appropriate fire and fuel management measures in line with predefined goals (e.g., reducing fire intensity or minimizing fire size).Fire and fuel management involves a comprehensive approach that integrates land management practices, considering fire regimes, impacts, values at risk, and multiple resource use activities. Despite the merits of each approach, limited resources often prevent the implementation of management measures in all areas with significant fire risk. Therefore, it is essential to prioritize high-risk areas and identify appropriate management strategies that maximize ecosystem protection. This study aims to develop an integrated method for identifying areas with the highest fire risk in the forests and rangelands of Guilan Province. The method combines simulation models of fire spread with landscape data (topography and fuel), fire weather conditions, and historical ignition point data. Additionally, anthropogenic factors, such as proximity to roads and settlements, are included in the fire risk modeling process, given their significant influence on fire behavior and regimes (Twidwell et al., 2016). The results of this research can support land-use planning, management, and future studies on fire hazards. Materials and Methods Study Area: This research was conducted in Guilan Province, which spans 1,404,400 hectares in northern Iran. The area ranges from -74 meters in the north to approximately 3,700 meters in the southeast. The province receives an average of 1,100 mm of annual rainfall, and the relative humidity averages 80%, decreasing during the summer months with 185 mm of rainfall during this period. Methods: Three Landsat-8 OLI/TIRS L1TP images from October 2021 were used to identify the main land cover types and produce a land cover map of the study area using ENVI 5.6 software. A map of surface fuel models was created by assigning local fuel models (Jahdi et al., 2020, 2023) and standard fuel models (Scott & Burgan, 2005) to the land cover types. Additionally, a vegetation density map and forest canopy density map (in percentage) for 2021 were generated from Landsat-8 images using the FCD model in ENVI software. Historical fire ignition data were used to calculate ignition probability by interpolating ignition points using the Inverse Distance Weighting (IDW) algorithm with a fixed radius of 5 km. The frequency of fire recurrence was determined from historical fire perimeters. Weather data, including temperature, relative humidity, wind speed, and wind direction, were collected for the fire season in the study area. To prepare input data for fire risk modeling: FlamMap 6 fire simulator (Finney, 2006) was used to model fire behavior, including fireline intensity and fire rate of spread, at a 100-meter resolution across the study area. Raster maps showing Euclidean distance from roads and settlements were created using ArcMap 10.8 software. Kernel density estimation was applied to convert the historical ignition point map into a fire density raster map. The fireline intensity, fire rate of spread, human factor maps, and fire history maps were rescaled to values between 0 and 1. These layers were combined to calculate the fire risk index (Xofis, Konstantinidis, Papadopoulos & Tsiourlis, 2020). The fire risk index values ranged from 0 (low risk) to 1 (highest risk). Results and Discussion Fifteen land cover classes were identified in the study area (Fig. 2, Table 4). The overall accuracy of the classification was 91%, with a Kappa statistic of 0.88. Dense broadleaf forest (30.6%) and orchard-irrigation farming (26%) were the dominant land cover types. Approximately 65% of the study area comprised natural forests, plantations, woodlands, shrublands, and rangelands. The remaining area included agricultural lands, built-up areas, water bodies, and bare lands. Based on the simulation outputs for fireline intensity and fire rate of spread (Fig. 4): In over 93% of the study area, the fire rate of spread did not exceed 0.25, and less than 7% of the area had higher values. Similarly, in over 90% of the area, the fireline intensity remained below 0.25, with about 10% exceeding this threshold. The highest values for fireline intensity and fire rate of spread were observed at high altitudes with dense vegetation, consistent with findings by Okano & Yamano (2015), which showed topography and slope significantly influenced fire behavior in broadleaf forests. Human factors, such as proximity to roads and the fire kernel density/fire history (Fig. 5), were found to impact a smaller proportion of the study area. This finding aligns with the study by Xofis et al. (2020). The integration of these components (Fig. 4 and Fig. 5) produced the final fire risk index map (Fig. 6). Fire risk values ranged from 0.0 to 0.68, with an average value of 0.11. Lowlands covered by non-burnable fuel models had the lowest risk (0.0), while the highest fire risk values were observed in the highlands with dense forest cover. Conclusion This study demonstrates how landscape information (topography and fuel), weather conditions, and historical fire data, combined with human factors, can be used to estimate fire risk spatially using fire behavior modeling. Moderate and high-risk areas account for 21% of Guilan Province, where fire management efforts should be prioritized. Specific measures include: Strengthening fire prevention infrastructure. Conducting fire prevention training to raise public awareness. Enhancing ecosystem resilience in vulnerable areas. While many fires remain random and difficult to fully control, estimating fire risk and developing prediction models can help mitigate fire hazards, reduce potential damages, and improve the allocation of fire prevention resources.