Examining the Transferability of Remote-Sensing Based Models of Live Fuel Moisture Content for Predicting Wildfire Characteristics

Live Fuel Moisture Content (LFMC) is a critical variable in improving fire risk estimations, which varies widely among different vegetation types and ecosystems. However, many regions do not have an operational LFMC mapping system, as such tools are difficult to develop. This raises the question of transferability, namely the potential of using LFMC estimates generated by existing models in regions they were not calibrated for. In this study, we examined the potential of three existing remote sensing-based models of LFMC, for estimating wildfire characteristics at a regional scale, using Israel as a case study. We tested two radiative transfer-based models (Australian and Global), alongside a machine learning-based model developed for the Mediterranean region. We compared the three models and found the Australian most suitable for Israel. Then, we conducted retrospective testing to analyze the effects of two variables derived from the LFMC estimates from the selected model. These factors were assessed with other fuel-related variables as predictors of wildfire characteristics including area burned, duration, and severity. Modelled LFMC were most strongly associated with burned area. The LFMC variables alone accounted only for 15% of the variability in burned areas. When additional fuel variables were included in the models, the proportion of variation explained in burned area increased substantially to 44%. We conclude that in regions characterized by high fuel heterogeneity, small fire sizes, and short fire durations there is a pressing need to develop new LFMC models using remote sensing data with better spatial and temporal resolutions.