Climate2Cat: Background and Overview

• Cat modellers build and use cat models to try and understand the physical damage impacts of many types of climate extremes. At the same time, climate scientists build climate models, and study climate variability and climate change.
• An obvious question is: could we use information generated by climate scientists to evaluate or adjust the cat models? Over the last 20 years or more, there have been various successful attempts to do that (see reference 1 for an example). It is, however, difficult, and is typically a research project every time. That’s because evaluating or adjusting cat models needs specific types of climate information, and that information is rarely generated by default in climate studies.
• This project is an attempt to create more efficient information pipelines from climate research to cat modelling. The goal is to encourage the creation of climate information in a form such that cat modellers (developers, or users) can use the information without having to do a research project, but could just plug it in. If we can achieve that goal, we can hopefully reduce the time from generation of the climate information, to application in cat models, from years to hours.
• To encourage the development of these information pipelines, this website defines, in detail, some of the basic types information that can be useful for evaluating or adjusting cat models, peril by peril. If climate scientists produce these types of information, then cat modellers should be able to use that information directly to evaluate or adjust their models. Climate scientists could make such information freely available, or charge for it…that’s up to them.
• There are many types of information that climate scientists might produce that could be relevant for cat modelling. We focus on the simplest and most easily usable information, which is information for evaluating or adjusting existing cat models. In that context, we describe some very specific examples. The examples we give describe information that would be of significant interest, right now, for many organisations that use cat models, but which is currently typically not available.
• This is a work in progress. We are starting by providing information about useful types of data for hurricane and other tropical cyclone models. Hail is coming soon, and after that other perils, depending on the level of interest.

Technical Details

Use of Event Frequencies

The easiest way to evaluate and adjust cat models is to evaluate and adjust the frequencies of events in cat model event sets. For making cat model frequency adustments, there are published algorithms (see reference 2). All our suggestions for presenting climate information are therefore based on providing information in terms of frequencies of events. For model adjustment, using frequencies is a general approach, as changes in frequencies can be used to specify changes in all other characteristics, without any real loss of generality. For instance, to specify an increase in the intensity of hurricanes (at constant overall frequency), you can specify an increase in the frequency of more intense storms, and a decrease in the frequency of less intense storms.

Rate vs GMST

Information about climate change should if possible be provided as present day rate of change versus global mean surface temperature (GMST), rather than as the change between period A and period B. When information is presented as a rate, it can easily be converted to a change between period A and B using past and future GMSTs, according to what future scenarios are of interest (see reference 3). For some perils and time-periods this rate-vs-GMST approach is not valid. An example is that rainfall in some regions is in some cases predicted to increase for a few years, and then ultimately decrease. For now, we will ignore those cases.

File Format

All data should be in csv files.

TC (Tropical Cyclones)

Basic Principles:

Information for tropical cyclones should be provided...
• ...for landfall frequencies, not basin frequencies. Otherwise the users have to do a research project to make the conversion. If there is interest we may be able to provide tools for this conversion.
• ...for a defined coastal region, such as the US hurricane states, or Florida, or Japan, or Kyushu, or Australia, or NE Australia.
• ...as a function of category of storm, using categories from -1 to 5, where -1 is ‘tropical depression’, 0 is ‘tropical storm’, and 1-5 are the usual SSHWS categories

Discussion of these Principes:

• Landfall vs basin information

  Providing information at landfall, as opposed to for basin storms, is essential for almost all cat modeling applications. Much climate research into TCs is presented in terms of basin rates, and that has been a barrier to using it in cat modeling. The conversion to landfall can be done statistically using historical data. It doesn’t have to be fancy. It doesn’t have to include any concept of how that conversion might vary with climate state, unless that’s information you want to communicate.

• Coastal Regions

  For basic tropical cyclone landfall rates, regions should be somewhat small-scale: US states or even counties, Japanese prefectures, Australian states. For changes in tropical cyclone rates due to forecasts or climate change, regions can be large-scale (US hurricane states as one region, Japan as one region, Australia as one region). For changes, regions can also be smaller-scale, if you believe you can identify results reliably at smaller scales, and if that’s information you want to communicate (small-scale might mean US states, Japan prefectures, Australia states)

• Dependence on intensity

  Specifying changes for cat03, and cat45, is not very helpful, because it creates an unrealistic jump between cat 3 and 4. It’s much better to smooth out the jump by providing information for each cat. That smoothing doesn’t have to be data-based, if it’s reasonable. One can argue that even using categories is slightly granular, and information could be provided at higher intensity resolution, e.g., by 5 mph wind-speed band, although one could debate whether that makes a material difference.

• Uncertainty

  A basic approach to providing TC frequency information is to specify frequencies (by cat, by region) as a single number, which is the mean of a Poisson distribution. This is generally adequate. However, a better approach is to specify a distribution for the mean of that Poisson distribution itself, to capture uncertainty. One simple option is to consider the mean of the Poisson to be from log-normal distribution, and specify two parameters (either mean and sd, or mu and sigma) which together define the distribution of the mean. The log-normal distribution has the benefit that it is easy to use. Another option is to consider the mean of the Poisson to be from gamma distribution, and specify the two gamma parameters. The gamma distribution has the benefit that the overall frequency distribution is a form of the negative binomial. At this point, not everyone would be able to incorporate log-normal or gamma uncertainty information into their calculations, although some would.

TC Example 1: Current Climate TC Landfall Rates

The frequency of landfalling TCs in the current climate is highly uncertain. There is great potential for different points of view as to a best assessment of that frequency, and as to the size and shape of the uncertainty around that best assessment. It is therefore of interest to cat modellers to obtain alternative views of current climate TC landfall frequencies. A plausible view would usually take into account how to use historical frequencies, account for past and present variability, and account for possible effects of climate change. Cat modellers could then compare such a view with the landfalling frequencies in the model they are using. If there is a difference, they may decide to use the alternative view as a sensitivity test, or to adjust the model to partially or fully reflect the alterative view.

Simplest case:

The simplest case would consist of a specification of current climate landfalling hurricane frequencies, as a Poisson mean, for the 7 intensity categories, by US state

Straightforward variations of the simplest case would include:
• the same, but by wind-speed instead of category
• the same, but by US county
• the same, but for other regions of the world.
• the same, but with uncertainty included around the Poisson mean (see the discussion about uncertainty above)
• the same, but by week or month during the year
More advanced variations of the simplest case might include:
• the same, but also by physical size of storms at landfall, perhaps by region, perhaps by cat
• the same, but by forward speed at landfall, perhaps by region, perhaps by cat

TC Example 2: Hurricane Seasonal Forecasts

Seasonal forecasts of hurricane frequency are of great interest to cat modellers. The idea here is to present seasonal hurricane forecast information in such a way that it can be used to adjust cat models in a straightforward way.

Simplest case:

The simplest case would consist of a seasonal forecast for US mainland landfalling hurricane frequency, presented as a Poisson mean (a single number, preferably not rounded to an integer).

Straightforward variations on the simplest case would include:
• the same, but by category, or wind-speed
• the same, but by Gulf vs East Coast vs Caribbean, or by state
• the same, but with uncertainty included around the Poisson mean (see the discussion about uncertainty above)

TC Example 3: Tropical Cyclones and Climate Change

Including the effects of climate change is of great interest to cat modellers. Many cat models already include the effects of climate change so far. But the effects of climate change are uncertain, and it is of interest for cat modellers to have access to different views. The idea here is to present an estimate of the current rate of change (versus GMST) of tropical cyclone properties due to climate change.

Simplest case:

The simplest case would consist of rate of change vs GMST of US landfalling hurricane frequencies, by category

Possible variations on the simplest case:
• the same, but by wind-speed instead of category
• the same, but by Gulf vs East Coast vs Caribbean, or other regions
• the same, but for other parts of the world
• the same, but for multiple climate models
• the same, but with uncertainty included around the Poisson mean, perhaps derived from multiple models (see the discussion about uncertainty above)

Hail

Resources

References

• S. Jewson (2023): The Impact of Projected Changes in Hurricane Frequencies on U.S. Hurricane Wind and Surge Damage: Journal of Applied Meteorology and Climatology

• S. Jewson (2023): A new simulation algorithm for more precise estimates of change in catastrophe risk models, with application to hurricanes and climate change: Stochastic Environmental Research and Risk Assessment // Free to view version

• S. Jewson (2021): Conversion of the Knutson et al. (2020) Tropical Cyclone Climate Change Projections to Risk Model Baselines: Journal of Applied Meteorology and Climatology // online software tool // Knutson et al. data // baseline conversion software source codes

Contact