Large ensemble flood loss modelling and uncertainty assessment for future climate conditions for a Swiss pre-alpine catchment.

Information on possible changes in future flood risk is essential for successful adaptation planning and risk management. However, various sources of uncertainty arise along the model chains used for the assessment of flood risk under climate change. Knowledge on the importance of these different sources of uncertainty can help to design future assessments of flood risk, and to identify areas of focus for further research that aims to reduce existing uncertainties. Here we investigate the role of four sources of epistemic uncertainty affecting the estimation of flood loss for changed climate conditions for a meso-scale, pre-alpine catchment. These are: the choice of a scenario-neutral method, climate projection uncertainty, hydrological model parameter sets, and the choice of the vulnerability function. To efficiently simulate a large number of loss estimates, a surrogate inundation model was used. 46,500 loss estimates were selected according to the change in annual mean precipitation and temperature of an ensemble of regional climate models, and considered for the attribution of uncertainty. Large uncertainty was found in the estimated loss for a 100-year flood event with losses ranging from a decrease of loss compared to estimations for present day climate, to more than a 7-fold increase. The choice of the vulnerability function was identified as the most important source of uncertainty explaining almost half of the variance in the estimates. However, uncertainty related to estimating floods for changed climate conditions contributed nearly as much. Hydrological model parametrisation was found to be negligible in the present setup. For our study area, these results highlight the importance of improving vulnerability function formulation even in a climate change context where additional major sources of uncertainty arise.

Flood risk (d)evolution: Disentangling key drivers of flood risk change with a retro-model experiment.

Flood risks are dynamically changing over time. Over decades and centuries, the main drivers for flood risk change are influenced either by perturbations or slow alterations in the natural environment or, more importantly, by socio-economic development and human interventions. However, changes in the natural and human environment are intertwined. Thus, the analysis of the main drivers for flood risk changes requires a disentangling of the individual risk components. Here, we present a method for isolating the individual effects of selected drivers of change and selected flood risk management options based on a model experiment. In contrast to purely synthetic model experiments, we built our analyses upon a retro-model consisting of several spatio-temporal stages of river morphology and settlement structure. The main advantage of this approach is that the overall long-term dynamics are known and do not have to be assumed. We used this model setup to analyse the temporal evolution of the flood risk, for an ex-post evaluation of the key drivers of change, and for analysing possible alternative pathways for flood risk evolution under different governance settings. We showed that in the study region the construction of lateral levees and the consecutive river incision are the main drivers for decreasing flood risks over the last century. A rebound effect in flood risk can be observed following an increase in settlements since the 1960s. This effect is not as relevant as the river engineering measures, but it will become increasingly relevant in the future with continued socio-economic growth. The presented approach could provide a methodological framework for studying pathways for future flood risk evolvement and for the formulation of narratives for adapting governmental flood risk strategies to the spatio-temporal dynamics in the built environment.

From global circulation to local flood loss: Coupling models across the scales.

Comprehensive flood risk modeling is crucial for understanding, assessing, and mitigating flood risk. Modeling extreme events is a well-established practice in the atmospheric and hydrological sciences and in the insurance industry. Several specialized models are used to research extreme events including atmospheric circulation models, hydrological models, hydrodynamic models, and damage and loss models. Although these model types are well established, and coupling two to three of these models has been successful, no assessment of a full and comprehensive model chain from the atmospheric to local scale flood loss models has been conducted. The present study introduces a model chain setup incorporating a GCM/RCM to model atmospheric processes, a hydrological model to estimate the catchment's runoff reaction to precipitation inputs, a hydrodynamic model to identify flood-affected areas, and a damage and loss model to estimate flood losses. Such coupling requires building interfaces between the individual models that are coherent in terms of spatial and temporal resolution and therefore calls for several pre- and post-processing steps for the individual models as well as for a computationally efficient strategy to identify and model extreme events. The results show that a coupled model chain allows for good representation of runoff for both long-term runoff characteristics and extreme events, provided a bias correction on precipitation input is applied. While the presented approach for deriving loss estimations for particular extreme events leads to reasonable results, two issues have been identified that need to be considered in further applications: (i) the identification of extreme events in long-term GCM simulations for downscaling and (ii) the representativeness of the vulnerability functions for local conditions.