|M.Sc Student||Jahshan Faris|
|Subject||Mapping Fire Risk Using Multiple Simulations of Fire|
|Department||Department of Civil and Environmental Engineering||Supervisors||PROFESSOR EMERITUS Maxim Shoshany|
|PROF. Yohay Carmel|
Understanding the spatial pattern of fire is essential for Mediterranean vegetation management. Fire risk maps are typically constructed at coarse resolutions using vegetation maps with limited capacity for prescribing prevention activities. This study describes and evaluates a novel approach for fire risk assessment that may produce a decision support system for actual fire management at fine scales for strategic planning actions.
FARSITE, a deterministic two dimensional fire growth and behavior model, using vector fire propagation technique based on Huygens’s principle, was activated to generate Monte Carlo simulations of fire spread events. FARSITE fuel models were adjusted for Mediterranean conditions. The study area was 200 km? of Mount Carmel, Israel.
All data layers were collected and processed from real information to construct a local GIS database including topographic data, climatic data, vegetation, and fuel models; In addition to human activity layer based on roads, hiking trails, and urban built up areas which served as the potential ignition map, 80% of the ignition sites were randomized from the human activity map, while the other 20% were randomly chosen from the Carmel wildland area.
The Monte Carlo simulation session consisted of multiple fire events. For each simulated fire event, a calendar date, fire duration, ignition location, climatic data and other parameters were selected randomly from the known distributions of these parameters. The resulting multiple fire spread maps of the randomized fire events were overlaid to produce an initial map of 'hotspots' of fire frequency; hereby the fire risk map.
The empirical distribution of fire events was Bootstrapped for results enhancement and statistical assessment, which means that sets of fire events were resampled with replacement from the “original” empirical distribution of fire events. These different sets were used to evaluate and enhance the risk map by calculating MSE and bias values of the Monte Carlo risk map.
The results revealed a clear pattern of fires, with high frequency areas. This pattern was compared to major documented historical fires showing a high degree of compliance. These results demonstrate the complexities of the fire behavior, showing a very clear pattern of risk level even at fine scales, where neighboring areas have different risk levels due to combinations of these factors.
The spatial pattern of the fire frequency map is affected by several factors such as fuel map, microclimate, topography and the distribution of ignition locations. Thus a sensitivity analysis was conducted to reveal the influence of each data layer on the final map product. These data layers were neutralized and the process of risk evaluation was activated for each neutralized set of data to evaluate its contribution to the final product by comparing it the reference risk map based on real data.
Few other spatial tests were also conducted to check if hotspots were related to any type of fuel model or other data; but no direct influence was detected. The results emphasize the complexity of fire behavior and fire risk mapping and exemplify the role of simulations and statistical approach to improve our understanding in this field.