A probabilistic risk assessment for the vulnerability of the European carbon cycle to weather extremes: the ecosystem perspective

Extreme weather events are likely to occur more often under climate change and the resulting effects on ecosystems could lead to a further acceleration of climate change. But not all extreme weather events lead to extreme ecosystem response. Here, we focus on hazardous ecosystem behaviour and identify coinciding weather conditions. We use a simple probabilistic risk assessment based on time series of ecosystem behaviour and climate conditions. Given the risk assessment terminology, vulnerability and risk for the previously defined hazard are estimated on the basis of observed hazardous ecosystem behaviour. We apply this approach to extreme responses of terrestrial ecosystems to drought, defining the hazard as a negative net biome productivity over a 12-month period. We show an application for two selected sites using data for 1981–2010 and then apply the method to the pan-European scale for the same period, based on numerical modelling results (LPJmL for ecosystem behaviour; ERA-Interim data for climate). Our site-specific results demonstrate the applicability of the proposed method, using the SPEI to describe the climate condition. The site in Spain provides an example of vulnerability to drought because the expected value of the SPEI is 0.4 lower for hazardous than for non-hazardous ecosystem behaviour. In northern Germany, on the contrary, the site is not vulnerable to drought because the SPEI expectation values imply wetter conditions in the hazard case than in the non-hazard case. At the pan-European scale, ecosystem vulnerability to drought is calculated in the Mediterranean and temperate region, whereas Scandinavian ecosystems are vulnerable under conditions without water shortages. These first modelbased applications indicate the conceptual advantages of the proposed method by focusing on the identification of critical weather conditions for which we observe hazardous ecosystem behaviour in the analysed data set. Application of the method to empirical time series and to future climate would be important next steps to test the approach.