Increased levels of greenhouse gases and the resulting warming on a global scale are expected to alter local climate- and weather-related risks [Field et al., 2012]. We are seeing evidence of this already: Drought-related wildfires, slow-moving hurricanes, and nuisance flooding are increasingly common events. To deal with these coming threats and challenges, scientists have investigated not only how severe these events might be but also how commonly they are likely to occur.
We propose a new strategy for providing this information: simulated stress testing for buildings and infrastructure to see how they stand up to extreme weather events. To do this, we need high-resolution climate information describing various aspects of local weather during extreme events.
This information can come from observing real-world phenomena or from simulations. Here we describe an approach that draws on both of these informational sources.
Mining Weather Forecast Archives for Information
Archived global operational weather forecasts can complement regional climate modeling and add a new source of information to impact studies. We suggest using such archives to generate a catalog of extreme events for stress testing.
We already know from reports and real observations when and where extreme events took place and what consequences they had for various locations. Archived global forecasts of recent events have even higher spatial resolution than for typical regional climate simulations. For example, archives from the European Centre for Medium-Range Weather Forecasts (ECMWF) provide forecasts with about 16-kilometer resolution, compared with 50-kilometer resolution from the Coordinated Regional Downscaling Experiment (CORDEX) [Gutowski et al., 2016]. Recent global operational forecasts also have a higher spatial resolution than the latest model reanalysis, but earlier events may have lower spatial resolution.
Climate Models and Their Limitations
Climate models simulate trends and events at a variety of scales, from global to local, but applying a chain of models to simulate events at local scale is not a trivial task. Regional climate models can provide a bridge between large and small scales; they are often used to downscale transient simulations from coarse global climate models to provide a more detailed description of the local climate change. However, global climate models may not necessarily reproduce the regional boundary conditions for extreme events, especially when their output is limited to a small set of simulations.
Climate modelers often construct simulations using multiple models, drawing on the strong points of each one. This approach is called ensemble simulation. Previous analyses of ensemble simulations show that on regional scales, because of internal climate variability, models can produce different outcomes for the same forcing, and these differences can be larger than differences in the models and in climate trends [Benestad et al., 2016; Deser et al., 2012]. Therefore, a small set of climate simulations provides an incomplete sampling with an inadequate description of the natural variability, which implies that this approach may not represent rare and extreme events well.
Even with perfect models and downscaling, generalizing from a small number of simulations is expected to provide misleading results (see “the law of small numbers” of Kahneman ). On the other hand, archives of downscaled results for large ensembles would require large volumes of data, and even then, there would be no guarantee that extreme events embedded in the results would be represented with sufficient realism. For example, climate models may neglect subtle but important aspects of air-sea coupling, which may affect the simulations of storms [Karnauskas and Zhou, 2018]. It is also a demanding job to sift through all the climate model simulations and find interesting events because the simulated weather is not synchronized with the real world.
To address these issues, the European Union project EU-CIRCLE, an initiative that works to strengthen infrastructure resilience to climate change, came up with a series of case studies that illustrated potential effects of a changing climate on various regions around the world.
Hurricane Matthew Visits Bangladesh
One of the EU-CIRCLE case studies looked into the potential effects of tropical cyclones on the city of Khulna in Bangladesh. This industrial city is located in an interior coastal area between the megacities of Dhaka and Kolkata. Climate change is expected to drive people away from inundated coastlines, and many of them could make their homes in Khulna. However, the city faces its own hazards from climate change: increased risk of cyclones and storm surges and the associated damage to roads, electrical utilities, water supplies, and other municipal systems.
Existing regional climate model results were insufficient to illustrate these hazards because it was difficult to find good specimens of tropical cyclones over the Bay of Bengal in the South Asia CORDEX regional climate model simulations. At present, there are few tropical cyclones in this region, and the driving global climate models may not sufficiently simulate conditions needed for cyclones to spawn. These simulations also have a spatial resolution of about 50 kilometers, which was inadequate for this case study.
As an alternative to the regional simulations, we asked, “What if a storm like X were to hit Khulna?” We chose forecasts for known tropical cyclones from different parts of the world and “displaced” them to Khulna, creating scenarios that EU-CIRCLE refers to as synthetic storms.
In this case, we selected Cyclones Nargis and Mora in the Indian Ocean (April–May 2008 and May 2017) and Hurricane Matthew in the North Atlantic/Caribbean (September–October 2016) and “moved” them to Khulna by changing the coordinates of the mean sea level pressure, 10-meter zonal and meridional wind components, and precipitation (Figure 1).
The EU-CIRCLE team met with critical infrastructure operators in Khulna in September 2017 to get input data for simulating storm impacts and then invited key stakeholders to a demonstration event in April 2018. The results showed that we could take city infrastructure information, incorporate it into our synthetic storm analysis, and visualize the effects of high winds and extreme precipitation on power distribution, roads, water supply, and the city’s health and education sectors.
Scaling Past Storms to Present Conditions
Such case studies can be used as a basis to extend the concept of synthetic storms through the use of statistical techniques. These techniques analyze the ways that a storm’s intensity, size, duration, and other factors depend on the global environment.
Historical storms can also be cataloged together with information about the ambient conditions, as well as a description of the storms’ consequences and damages. This makes it possible to exploit all relevant information for stress testing under new climate conditions. For example, what would be different if Hurricane Andrew had occurred in 2018 rather than 1992?
Using the Method Effectively
The synthetic storm approach provides a wealth of information, but users must be made aware of the strengths and limitations of the technique. This method cannot be used for all types of climate change adaptation. Therefore, it is essential for users to interpret the information correctly and understand these limitations.
Synthetic storms involve a preselection of specific events. This approach differs from picking random samples, which is needed for proper statistical analysis such as return value analysis, which indicates how long a time there typically is between each event. Because extreme events occur relatively infrequently, a selection of the most extreme events represents a high, albeit unknown, return value. Because the selection is made from many locations and hence would involve multiple return values, one limitation is that it is not possible to say much about probability or return values.
There may also be a physical consistency that depends on local factors, such as topography, and it is unrealistic to move storms everywhere. However, the synthetic storm strategy may be particularly suitable for coastal cities and island states surrounded by oceans.
Despite its limitations, this “shortcut” approach can be valuable to decision-makers who do not need estimates of probabilities and return values but merely a plausible estimate for storm magnitude or intensity. For instance, planning for disaster risk reduction and security often relies on perceived perils, and in this context, such stress testing may provide a valuable contribution.
This work has been supported by the Norwegian Meteorological Institute and EU-CIRCLE (grant 653824).