Extreme value modelling of storm damage in Swedish forests

Abstract

Forests cover about 56% of the land area in Sweden and forest damage due to strong winds has been a recurring problem. In this paper we analyse recorded storm damage in Swedish forests for the years 1965&ndash;2007. During the period 48 individual storm events with a total damage of 164 Mm³ have been reported with the severe storm on 8 to 9 January 2005, as the worst with 70 Mm³ damaged forest. For the analysis, storm damage data has been normalised to account for the increase in total forest volume over the period. <br><br> We show that, within the framework of statistical extreme value theory, a Poisson point process model can be used to describe these storm damage events. Damage data supports a heavy-tailed distribution with great variability in damage for the worst storm events. According to the model, and in view of available data, the return period for a storm with damage in size of the severe storm of January 2005 is approximately 80 years, i.e. a storm with damage of this magnitude will happen, on average, once every eighty years. <br><br> To investigate a possible temporal trend, models with time-dependent parameters have been analysed but give no conclusive evidence of an increasing trend in the normalised storm damage data for the period. Using a non-parametric approach with a kernel based local-likelihood method gives the same result.