A novel framework for fire risk assessment in Kazakhstan: integrating machine learning and remote sensing

Wildfires present a significant threat to ecosystems, property, and human life in Kazakhstan. Understanding fire hazards is essential for effective management and mitigation of these risks. This study develops a comprehensive fire hazard index for Kazakhstan by integrating static, long-term landscape factors with dynamic, real-time weather and vegetation conditions. The static component employs a machine learning approach, specifically the Random Forest algorithm, trained on a dataset that includes topographic variables derived from the SRTM DEM, land cover classifications from MODIS Terra/Aqua LULC products, and historical fire occurrence data from NASA FIRMS. This model quantifies the inherent fire susceptibility of various landscapes based on these enduring characteristics. The dynamic component captures short-term fluctuations in fire risk by incorporating satellite-derived vegetation information and meteorological observations. The MODIS-derived Normalized Difference Vegetation Index (NDVI) serves as a proxy for fuel availability and moisture content. Spatially interpolated weather data such as temperature, humidity, wind speed, and precipitation provide the necessary meteorological context. The dynamic index is calculated using a modified Canadian Fire Weather Index (FWI) system, specifically adapted to account for the influence of live fuel moisture, as indicated by NDVI, on fire ignition and spread dynamics. The final fire risk index is created by additively combining the static and dynamic components, offering a spatiotemporal perspective on fire risk. This integrated approach allows for the assessment of both the underlying susceptibility of a landscape to fire and the immediate effects of weather and vegetation conditions. The resulting high-resolution fire hazard maps are intended to inform fire management decisions, optimize resource allocation for fire prevention and suppression efforts, and support targeted interventions in high-risk areas. This research underscores the value of combining machine learning techniques with remotely sensed data for enhanced fire risk assessment in Kazakhstan, facilitating more proactive and effective fire management strategies.