Validating expert-based fuel model by field observations and simulations

Accurate fuel model mapping is essential for supporting the prediction of forest fire ignition and fire propagation. Although standardized fuel model classifications are widely applied in fire sciences, their performance is often limited when evaluated against field observations, largely due to the high variability within the different fuel categories. Especially in Central Europe there are less experiences with the application of different fuel model classification due to the lack of experiences in the predicting fire behavior under the specific environmental conditions and the lower number of larger fire events. This study addresses these needs by proposing a validation framework to ensure that fuel models assigned to a certain forest patch or landscape allow to represent real-world fire behavior. 

To develop the fuel model map for this study, experts combined field measurements on fuel loads with the results of the interpretation of aerial imagery to classify fuels, assigning classes for each 10x10m pixel according to the Scott and Burgan (2005) fuel models based on their interpretation. The proposed validation framework of the fuel model map for this study integrates observed field data from forest fires and prescribed burns in the past to estimate selected fire behavior parameters, such as flame length and rate of spread (ROS). These field observations serve as a ground truth to evaluate the accuracy of a developed customized fuel map using expert-based knowledge. Additionally, we simulate fire behavior with the BehavePlus package for the expert-assigned fuel models, to determine if the simulated parameters match the observed field data, thereby validating whether the fuel model assigned to a given area is both appropriate and provides physically realistic fire behavior. Furthermore, we utilize the Rothermel R package, which implements the mathematical equations of the Rothermel (1972) fire spread model, to reverse-analyze field data and identify the most probable fuel model for a given condition. In a next step, we compare the fuel models suggested by the algorithmic with the fuel models assigned by the expert judgments and the fire behavior parameters derived from BehavePlus. 

The results of this study show that customized fuel models based on expert knowledge outperform standardized fuel classifications in representing real-world fire behavior. Reverse fitting of field data using the Rothermel’s model is likely to show differences between algorithmically derived parameters and expert-assigned fuel models, particularly in complex and heterogeneous landscapes. Overall, the integration of field observations with expert-based fuel modeling is expected to reduce uncertainty in fire behavior simulations by: i) comparing simulated fire behavior parameters to field observations; and ii) using the Rothermel R package to validate expert-assigned fuel models, diagnose mismatches and refine fuel assignments.