Skyfora updates AI-powered tropical cyclone model, gets ILS adoption
Skyfora, a Helsinki-based weather intelligence startup that utilises artificial intelligence to derive tropical cyclone intensity forecasts, has announced an updated model and also noted that it has on-boarded a customer from the insurance-linked securities (ILS) market.
Skyfora believes its model forecasts can provide a new source of enhanced insights into potential hurricane landfall impacts for the insurance, reinsurance and insurance-linked securities (ILS) market.
Today, Skyfora has announced an updated version of its Tropical Storm Seasonal Forecast model which it claims has proven skill, particularly for the important Gulf of Mexico area.
The model provides probabilistic predictions for the likelihood of various events in different categories for the upcoming tropical cyclone season.
This covers tropical cyclone (TC) intensity, storm genesis by region, and storm landfall by region, while Skyfora also provides its clients with forecasts in early April, May and June for the Atlantic basin, it also publishes mid- season forecasts in July, August and September, according to its customer needs.
Skyfora’s model uses Bayesian neural networks and probabilistic machine learning to its full extent, with the model trained using several data sources including historical atmospheric, land and oceanic climate variables.
Skyfora claims that cross-validation experiments against the baseline show a mean absolute prediction error for major hurricanes is reduced by 50% against the baseline, and the Gulf of Mexico landfall prediction has a correlation of over 0.7 with the true values for years 2011-2021.
“Seasonal forecasting and particularly skilled landfall forecast have been traditionally difficult to develop and prove due to the chaotic nature of tropical cyclones,” explained Dr. Svante Henriksson Founder and CEO of Skyfora. “Last year we published our first seasonal forecast, but since then we have changed focus from traditional point predictions and decided to deploy Bayesian deep learning instead, as the resulting probability distribution is more useful for our re-/insurance, catastrophe modelling and ILS customers. This has proven to be the right decision.”
The company is new to the modelling space, but is already getting traction for some in insurance-linked securities (ILS).
New technological approaches to weather and climate modelling are being very important to ILS managers and Skyfora’s Tropical Storm Seasonal Forecast model is already being utilised by ILS investment manager Securis Investment Partners in London.
“Skyfora’s team is ambitious, and they provided a new approach to a very challenging problem. By working closely together we’re able to focus the work on the most impactful metrics and most relevant questions of the business. By monitoring how the forecasts, and risk, evolve during the season we hope Skyfora will help us more accurately and more dynamically manage our risks,” Dr Paul Wilson, Securis’s Head of Non-Life Analytics explained.
As new technologies become more readily-available, the scientific inputs into ILS investment decision-making are going to become richer and more ubiquitous, with new generation risk modelling techniques a key development for the ILS market.