Early warning meteorological fire danger over Central Europe

Meteorological fire danger has been steadily increasing over Europe in the last decades, not only over the Mediterranean South that is recurrently affected by extreme fire weather and where the largest fire events take place, but also over Central, Eastern and Northern countries that are facing more and more events. The two most recent examples are the devastating fires in Rhodes and northern Greece in 2023, and those in France, Spain, Portugal, Slovenia and Czechia in 2022 when the total of burnt area almost reached the record value of 2017. The increase in severity of fire events is of major concern for all European countries, but special attention should be devoted to Central Europe where large fires, usually driven by the compound effect of droughts and heatwaves (e.g., 2018, 2022), are posing new challenges at the levels of fire management and fire forecasting. We present a statistical model of energy released by wildfires that allows calibrating the Canadian Fire Weather Index (FWI) over three major land cover types (forest, shrub, and agriculture) covering an area encompassing Central Europe (3.5-17E and 45-62N). The model consists of a doubly truncated lognormal body distribution with generalized Pareto tails (DaCamara et al., 2023) that incorporates FWI as a covariate of its parameters. For each land cover type, the model is fitted to the set of observed values (from 2001 to 2022) of the logarithm of Fire Radiative Power associated to hotspots as detected by the MODIS instrument on-board Terra and Aqua platforms. For each model, goodness of fit is evaluated by using the Anderson-Darling test to assess the strength of the evidence against the null hypothesis that the sample follows the distribution. The fitted models allow estimating for each land cover type the probability of exceedance of a predefined threshold of log(FRP) for each day and grid point. Five classes of fire danger (low, moderate, high, very high, and extreme) for each land cover type are then defined by analyzing the spatial and temporal variability of the distribution of pixels among classes as well as the distribution among classes of FRP associated to hotspots, such that classes of higher fire danger tend to concentrate in the fire season, and fires with high values of FRP occur in pixels classified in the classes of high, very high and extreme danger. The procedure is further validated by examining several case studies that were chosen because of unusually intense fire events or because of the high number of occurrences. This work was supported by EUMETSAT Satellite Application Facility on Land Surface Analysis (LSA SAF) and by Instituto Dom Luiz (IDL), a research unit financed with national funds (PIDDAC) by FCT (UIDB/50019/2020). References: DaCamara, C. C., Libonati, R., Nunes, S. A., de Zea Bermudez, P., Pereira, J. M. C. (2023). Global-scale statistical modelling of the radiative power released by vegetation fires using a doubly truncated lognormal body distribution with generalized Pareto tails. Physica A: Statistical Mechanics and Its Applications, 625. https://doi.org/10.1016/j.physa.2023.129049