Remote sensing-based adaptation of the Canadian Fire Weather Index towards developing a new fire danger monitoring system for Europe

Kurzfassung

Wildfires are one of the most critical natural hazards for society. Especially as climate change is aggravating environmental conditions and leading to a higher number of large-scale wildfires, monitoring fire danger is of great importance for raising awareness of the population and taking mitigating measures accordingly. Different methods and ratings of fire danger integrate components of fire danger such as atmospheric and climate conditions and the availability of combustible material of dead or live vegetation. The Canadian Fire Weather Index (FWI) system is a widely used, best-performing fire rating system integrated into the European Forest Fire Information System (EFFIS). It is a meteorological fire index developed for a standard fuel type in Canada and therefore doesn't account for vegetation and forest type variations. This thesis develops and tests a new approach for overcoming those limitations by integrating the remote sensing-based vegetation index Vegetation Condition Index. The study tests whether integrating vegetation indices into the FWI system improves the fire danger estimation and evaluates the forecast based on detected burned area and burn severity. The fire danger index is calculated based on ERA5-Land climate reanalysis data using ERA5 historical fire danger indices as reference. Integrating the Vegetation Condition Index (VCI) based on Sentinel-3 (S3), the FWI is adapted directly by systematically testing different factors for incorporating the VCI. The study was conducted monthly for the fire season from June to September for the time series from 2016 to 2021 with application to the Spanish region of Andalusia. The performance of fire danger indices were analyzed with Pearson's correlation coefficient r and the classification metrics accuracy and Intersect-over-Union (IoU) based on a random forest model. This thesis found that the VCI integration of the higher resolution data of S3 resulted in an improved spatial differentiation of the fire danger assessment based on ERA5-Land. The study on parameter tests was based on two basic scenarios for decreasing and increasing the fire danger rating. The findings suggest a possible underestimation of the fire danger based on the original FWI since the adapted, increased fire danger index resulted in slightly better correlation coefficients and classification metrics. While accuracy showed as a poor comparative metric, the IoU score for unburned areas was the most informative statistical parameter since it revealed parameter combinations in which the number of False Negatives or false alarms of burned areas decreased in favor of True Negatives. Forecasted fire danger and burn severity showed a low to mid-range correlation. The parameter tests changed mainly the type of linearity, with an increase of the FWI shifting towards positive correlation and a decrease to stronger negative correlation. The monthly analyses showed better results the higher the number of samples and indicated only little seasonal variability. Based on this work, further research should follow to analyze the impacts of the modifications and integration of the vegetation index into the FWI more in-depth.