Statistical calibration of probabilistic medium-range fire weather index forecasts in Europe

Abstract. Wildfires are increasing in frequency and severity across Europe, which makes accurate wildfire risk estimation crucial. Wildfire risk is usually estimated using meteorological based fire weather indices such as the Canadian Forest Fire Weather Index (FWI). By using weather forecasts, the FWI can be predicted for several days and even weeks ahead. Probabilistic ensemble forecasts require verification and post-processing in order to provide reliable and accurate forecasts, which are crucial for informed decision making and an effective emergency response. In this study, we investigate the potential of non-homogeneous Gaussian regression (NGR) for statistically post-processing ensemble forecasts of the Canadian Forest Fire Weather Index. The FWI is calculated using medium range ensemble forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) with lead times up to 15 days over Europe. The method is tested using a 30 day rolling training period and dividing the European region into three training areas (Northern, Central and Mediterranean Europe). The calibration improves FWI forecast particularly at shorter lead times and in regions with elevated FWI values i.e. areas with a higher wildfire risk.