Climate Change Loss Impact
In this section we look at various studies that investigate the potential impact on insured losses from the projected climate change scenarios.
Eurowind™ Climate Change Stress Tests
As discussed in the section CoreLogic Study: Storm Frequency Changes, both SRES forcings led to a an increase in the frequency of Daria strength events by the end of the 21st century (Figure 4). On this basis, a climate change-specific sensitivity experiment was performed using Eurowind™ to investigate the impact of doubled frequencies of Daria-strength storms on modelled losses. The impact is shown in Figure 13 in the format of loss exceedance curves for all 24 Eurowind™ countries, in addition to individual curves for the UK, France, and Germany.
Figure 13 Changes in the Eurowind™ Loss Exceedance Curve due to doubling the frequency of all storms of the stochastic event set reaching and exceeding the severity of Daria (Clockwise from Top Left, All Countries, United Kingdom, France, and Germany). Source: CoreLogic
Depending on the country, the projected changes in losses due to the doubling of frequencies of storms of at least Daria strength are in the 10% to 20% range between the 100- and 200-year return period regions. This is of interest to insurers as Storm Daria sized events are well within the capital allocation and solvency requirements but could have a significant impact upon conducting day-to-day business, such is their potential for disruption. Events of this magnitude are also likely to have a demand surge component thus exacerbating the financial impact of the storm.
Figure 14 Impact on European windstorm loss as a function of global temperature rise as estimated by seven studies from the scientific literature. The top panel shows results from each study, whilst the lower panel shows all results combined. The marker sizes in the top panel represent the weight assigned to the model, . Source: Ranson et al 2014, Figure 2.
Further CMIP5 Loss Impact Studies
CMIP5 models have been widely used in scientific literature and there are several significant studies of the impact of climate change on European windstorm loss using CMIP5. For example, Ranson et al. (2014) provide a synthesis of previous studies, a summary of which is depicted in Figure 14. Future warmings of 2.0° and 2.4° (i.e., mid-century SSP5-2.4 & SSP5-8.5 conditions) would lead to increases in losses of 18.4% and 22%, respectively. However, as can be seen from the histogram shown in Figure 14, there is a large uncertainty associated with this prediction. Furthermore, these studies have been superseded by more modern CMIP6 studies.
CMIP6 studies
CMIP6 outputs enable us to develop better insights into the impact of climate change on insured losses as the underlying modelling is at a higher resolution and resolves the physical processes better than the CMIP5 outputs.
A recent study from Little et al 2023 utilised a total of eight CMIP6 models, the results of which are summarised in Figure 15. Theoretically, the results of this study are more robust than those summarised in Ranson et al 2014, owing to the improvements in the horizontal resolution of CMIP6 models.
The impact of SSP2-45 and SSP5-85 were modelled over a time horizon at the end of the 21st century, 2070–2100, with respect to a historical period spanning 1980-2010. Impact was assessed in terms of storm severity using an SSI based on the near surface maximum windspeed of each modelled event.2
The SSI results are summarised in Figure 15, which shows changes across the whole European domain, northwest Europe (NW), northeast Europe (NE), and south Europe (S). The authors consider the effect of adaptation measures (labelled ‘AD’ in the figure). Here, we concentrate on the worst-case of no adaptation (labelled as ‘NAD’ in the figure). When compared to CMIP6 historical simulation, increases occur in both NW and NE Europe, whilst there is a decrease in South Europe, consistent with the tripolar hazard signal discussed above.
Using the method discussed previously that utilises SSP2-4.5 at 2100 as a proxy to SSP5-8.5 at 2050, we can model results centred on 2050. The result of this approximation provides All Europe category SSI increases by 43.7% in SSP5-8.5 2050 with a SSP2-4.5 change of 11.2%. Similarly, the NW Europe SSP5-8.5 2050-centred change represents a ~50% increase.
There is, however, large uncertainty associated with these mean estimates, driven by the spread of multi-model projections. A single model within this ensemble, BCC-CSM2-MR, has the largest increase in track density over NW Europe, and if removed from the analysis, reduces the SSP5-8.5 end-of-century change in NW Europe to +13.3%. Assuming a linear dependence of damage with respect to temperature change, according to Figure 3, this implies an increase of just 7% under SSP5-8.5 centred on 2050.
2 The study also includes estimates of the impact on loss, where the latter is approximated by weighting the SSI by a time-dependent estimate of the local population at each model grid point. Projected changes in population dominate the results.
Figure 15 : impact on the European region, NW Europe (45°N–71°N, 23.5°W–13°E), NE Europe (45°N–71°N, 13°E–31°E), and S Europe (35°N–45°N, 23.5°W–31°E). Source: Figure 4 of Little et al. 2023
A further recent study by Severino et. al., (2023) combined 30 different CMIP6 GCMs to assess the possible impact of climate change on European windstorm damage under SSP585 on a time horizon of 2070-2100 (with changes relative to a historical period of 1980-2010). Their results are reproduced in Figure 16 in terms of a spatial map of Average Annual Damage (AAD) in the left panel, and by boxplots showing AAD along with changes in yearly and 15-yearly return period (RP) damage. Losses were calculated using the CLIMADA model (Aznar-Siguan and Bresch, 2019), using exposure from the ‘produced capital layer’ of the LitPop data set (Eberenz et al 2020), and a single vulnerability curve from Schwierz et al 2010, which does not differentiate between lines of business.
The median change in AAD is positive in five of the regions shown in in Figure 16: British Isles (BI, +20%), Western Europe (WEU, +34%), Central Europe(CEU, +9%), the Mediterranean 380 and Balkan region (MED, +1%), and Scandinavia (SC, +8%). The remaining two regions exhibit negative changes: the Iberian Peninsula (IP,−7%), and Eastern Europe (EEU, −14%).
Once again, the positive increases in the areas of significant interest for insurance purposes, i.e., Western Europe, the British Isles, and Central Europe are consistent with the tripolar signal in track density changes discussed above.
The study does not include estimates either of the impact on the timescale horizon centred on 2050 or of less extreme SSPs. Once again, assuming a linear dependence of damage with respect to temperature change, according to Figure 3, SSP5-8.5 centred on 2050 would lead to: British Isles (BI, +11%), Western Europe (WEU, +19%).
Figure 16 British Isles (BI), the Iberian Peninsula (IP), Western Europe (WEU), Central Europe (CEU), the Mediterranean and Balkan region (MED), Scandinavia (SC), and Eastern Europe (EEU). Source: Figure 17 of Severino et al. 2023
There is, however, large uncertainty associated with the projected changes. For instance, in the map shown in Figure 16, more than 75% of the GCMs agree on the sign in the change in AAD in the regions shown with hatchings. Conversely the areas without hatchings exhibit greater uncertainty with less than 75% of the GCMs agreeing on the sign. It is in these regions where the majority of insured exposure is concentrated .
Similarly, the boxplots reveal a large spread, with the 25th percentiles reaching large negative values in regions of high insured exposure e.g. BI, WEU categories.
Severino et al. 2023 also contains a sensitivity analysis to determine which factors dominate this uncertainty in the projected changes in damage. Across all regions and for each of the three damage-metrics considered, the uncertainty is dominated by model error i.e., by the choice of CMIP6 model. Natural variability can be estimated as the variability exhibited by an ensemble of runs made using each model separately. The contribution of natural variability to the damage uncertainty was found to be much smaller than that of model error.