Challenges in using event storylines for climate risk assessments. The example of a severe landslide event in Austria.

Event storylines are a variant of storylines and can be used to explore the consequences of low-likelihood high impact events. In particular, they provide the basis for a realistic emergency operations center exercise: stakeholders and scientists can run through different - also management - scenarios and asseess their complex and cascading risks and costs.

Different approaches to implementing event storylines exist; most are based on some variant of the pseudo global warming approach: an observed event is simulated under the actual (boundary) conditions and under modified boundary conditions representing selected scenarios. The simulations can then be fed into quantiative impact models and further be used for qualitative assessments. But this (in theory) very elegant approach comes along with several challenges in its practical implementation.

Here we use the example of a severe landslide event in Southern Austria to illustrate these challenges and present solutions. Heavy rainfall, caused by a slowly moving cut-off low and falling on saturated soils, triggered at least 952 landslides and resulted in substantial damage of infrastructure and buildings. To model the event, we combine kilometer-scale regional climate model simulations with a statistical landslide model, trained on a comprehensive dataset of observed landslide, meteorological, geological, topographical and vegetation data. We simulate the event under present, observed boundary conditions, as well as under modified conditions representing different global warming levels as simulated by global climate models. The actual implementation of changes in boundary conditions, however, is not a priori clear. Also, even though the event is well simulated, it is dislocated compared to observations by a few tens of kilometers. This dislocation is of the same order of magnitude as the area affected by landslides and thus makes a direct use of the simulations for driving the landslide model unfeasible for representing a specific event. 

In a set of sensitivity studies, we first explore the influence of (1) simulating the event with climatological boundary conditions over a large domain with spectral nudging vs. event-type specific boundary conditions over a small domain without spectral nudging; and (2) imprinting altitude dependent or constant changes in different atmospheric variables, from temperature only to temperature, humidty and sea level pressure. The results depend strongly on the implementation. However, a process-based analysis reveals that only the small-domain variant with sea level pressure and consistent altitude-dependent changes in temperature and relative humidity simulates physically plausible changes. Second, we develop a delta change approach, which (1) replaces temporal by spatial averaging to calculate change factors, and (2) applies changes separately to precipitation at different time-scales.  Finally, we discuss the relevance of carefully defining the event in time, including preconditioning by antecedent precipitation which may change in a warming climate, and how changes in these preconditions can be simulated.   

Our study demonstrates the great potential of event storylines for risk assessments, but also highlights the need for a range of critical choices and post-processing steps that need to be carefully considered to arrive at plausible results.