- Opportunity to drive science forward for extra-tropical cyclones, the biggest and most disruptive risk to the UK and much of Europe.
- Learn data science skills and use cutting-edge seasonal forecasts (e.g. SEAS5).
- Engagement with (re)insurance sector, thus results with real world impact and enhanced post-PhD job prospects. A placement/internship is envisaged.
Severe extratropical cyclones (ETCs) are one of the largest risks on the balance sheet of insurance companies (Hillier, 2017), and can cause flooding (coastal, pluvial) and/or wind damage (e.g. De Luca et al., 2017; Donat et al., 2011; Lavers et al., 2011). This impact varies between years. In Europe, particularly the UK, ETC occurrence and characteristics are to be related to known factors such as the North Atlantic Oscillation (NAO, e.g. Walz et al., (2018)). This subject is of particular current interest because those factors (e.g. NAO) are becoming predictable on seasonal time scale (e.g. Scaife et al., 2014). Further, the latest research has demonstrated forecast skill in state-of-the-art seasonal forecasts for severe wind storm events capable of damage, and that the NAO cannot be the only controlling influence (Befort et al., 2018). Thus, this project will focus on a hindcast approach (e.g. a forecast of the 2015-6 winter as if made in October 2015) to estimate the potential impact on business decision making of seasonal forecasts.
The project will address this topic by investigating the following research questions:
What meteorological variables or metrics (e.g. NAO, storm count) give predictability to estimates of loss in the upcoming season?
When should these measures have attention paid to them? E.g. do extremes of the NAO provide skill, whilst moderate NAO phases do not (Renggli et al., 2011)? This will develop the growing understanding of the conditions in which predictability is possible.
Where (i.e. geographically) is prediction possible?
Why is the metric valid? i.e. how is explained in terms of physical processes? This underpinning understanding is essential to give comfort to decision makers (e.g. that this isn't just a statistical artefact), and it gives the basis for insights into another key consideration i.e. what is the uncertainty in the predictions?
These will feed into practical considerations:
In what years would additional information about the season ahead have made most difference to reinsurance decisions?
Should (re)insurers align their annual renewals to immediately prior to a season (e.g. Oct-Mar) such that the information available to them is maximized?
A core of the work is low risk, but scope exists for a student to innovate and excel (e.g., understanding the predictability of spatial correlations between countries in Europe or between regions such as Europe and N. America). Elements of the proposed analysis are:
- Literature review.
- Use of historical SEAS5 seasonal forecast data (ECMWF).
- Hypothesis driven investigations of observable quantities (e.g. storm counts), and the physical processes driving any predictability. Methods (e.g. storm track detection, back-tracking of relevant air parcels) will be selected as appropriate.
- Review, perhaps qualitative, of decision making in last 10-15yrs within one or several (re)insurers to understand this process. To include a focus on any differences if the seasonal forecasts now available would have been used in the past.
- Construction and running of a 'toy' illustrative decision-making tool so that the impact (e.g. financial) of additional information about the season ahead can be quantitatively commented upon.
Training and Skills
Specifically for this project, the PhD student will gain state-of-the-art data science skills applicable to both meteorological data and financial losses; such large datasets are sometimes referred to as ‘Big Data’. Training will be provided in GIS techniques, approaches used in atmospheric science, analytical methods (e.g. in R). To gain on-the-ground experience of (re)insurance, it the student will have the opportunity to undertake placement(s) in markets such as London and attend industry conferences (e.g. Impact Forecasting). Training will also be available in catastrophe model design and use, with a focus on flooding and wind models.
Year 1: The student will familiarise themselves with state-of-the-art seasonal forecasting, including SEAS5 hindcasts and their numerical interrogation, with the aim of identifying candidate metrics and physical processes with promise to be skilful predictors for further investigation. In parallel a review, perhaps qualitative, of decision making in last 10-15 ETC seasons within one or several (re)insurers will be undertaken.
Year 2: Hypothesis driven investigations of observable quantities (e.g. storm counts), and the physical processes driving any predictability.
Year 3: Whilst continuing to research atmospheric science, the main addition this year will be to construct and running of a 'toy' illustrative decision-making tool so that the impact (e.g. financial) of additional information about the season ahead can be quantitatively commented upon.
Depending upon progress, the scope of the project may expand to include (i) correlations between flooding and wind damage or (ii) the decline in European storminess since the year 2000, and could this have been noticed coming with the hindcasts. The (re)insurance industry is keen to understand the answers to both of these.
Partners and collaboration (including CASE)
This project is co-designed with and supported by the Lighthill Risk Network, a Level-1 (i.e. top tier) CENTA partner, who will supply both (i) expert knowledge of the (re)insurance sector and (ii) supervisory input into the project. Via this partner, we will secure further interest and input from additional interested relevant partners (e.g. Lloyd's of London, Aon) from the insurance industry. This will be further discussed and clarified once the project has started, depending on the skill set of the applicant.
For information about this project, please contact Dr John Hillier (firstname.lastname@example.org).