Seasonal climate impact prediction

By James Done, C3WE Science Lead and Willis Research Fellow

This guest blog is reproduced with kind permission from Willis. The original post can be found here in Willis Wire November 2014.

We couldn’t possibly have anticipated Hurricane Sandy in advance of the season, or could we?

Seasonal climate forecasts have been produced routinely for many years yet their uptake by the industry has been slow. Given that insurers and reinsurers live and die by high-impact weather and climate events why is this?

For too long climate scientists have been guilty of generating terabytes of climate forecast data, throwing it over the fence, and expecting it to be picked up and used. Even worse, it is barely noticed. There remains a huge gulf between what the industry needs and what scientists produce. A forecast for increased likelihood of above average U.S. temperatures, for example, is of little value to an industry that lives primarily at the extremes. A transformation is needed of the way both the industry and climate scientists conceptualize seasonal climate forecasting.

Why Forecasts Beyond 7-10 Days Are Unreliable

Seasonal prediction of specific events will never be possible. Before exploring the requirements for useful, usable and actionable seasonal climate forecasts it is helpful to review the current state of the science and current industry application. Everyday weather forecasts derive skill from accurate knowledge of weather at the start of the forecast and project forwards in time using the physical laws of the atmosphere. The reason forecast skill plummets beyond 7-10 days is that unknowable small-scale details grow over time to overwhelm the forecast. This begs the question, if we can’t predict the weather beyond 7-10 days how can we make skilful seasonal climate forecasts?

Seasonal Climate Forecasts Are More General

Skill at longer timescales derives from knowledge of the more slowly evolving components of the climate system and their influence on the distribution of weather. Rather than forecasting the details of day-to-day weather, seasonal climate forecasts tell us about the weather over an entire season such as average temperature and aggregate rainfall totals. 

Just like we can predict that summer will be warmer than winter due to a predictable increase in the incoming energy from the sun, we can say something about the upcoming season due to predictable changes in the slowly varying climate. This predictability arises from persistence or memory in the climate, known climate cycles such as El Niño, and extrapolation of climate trends. 

Likelihood statements that a season will see above- or below-normal seasonal total rainfall or season average temperatures are a far cry from societal need for guidance on the timing, location and scale of high-impact weather.

The bad news is that seasonal prediction of specific events is not possible today. Even worse, it will never be possible. A recent peer-reviewed study by the Willis Research Network showed that a North Atlantic hurricane season can become twice as active purely due to small-scale processes that are unknowable in advance of the season. This means that even if we had complete knowledge of pre-season conditions and a perfect forecast system we will never be able to forecast perfectly the numbers of hurricanes every season, let alone specific landfalling locations.

These unknowable small-scale processes are one reason why we should still expect forecast busts, such as the much anticipated 2013 hurricane season that turned out to be one of quietest in recorded history. But as the science improves forecast busts should occur less often and the average investment return over a number of years should remain positive.

Advances in Forecasting

Despite this barrier to predicting the specifics of a season, there lies a rich middle ground between what is desired and what is possible that remains largely unexplored.

Given that the likelihood of weather extremes (such as flooding rains, extra-tropical windstorms and tropical cyclones) changes with the background climate state, a forecast deviation in the background state has the potential to also say something about the deviation in likelihood of high-impact weather events. Already, some centers issue seasonal forecasts of weather extremes (tropical cyclones being the most well known), but there remains huge potential for seasonal forecasts of other weather and climate hazards.

Perhaps the greatest advance will be made through a change in forecasting the weather and climate to forecasting the impact in industry-relevant terms, such as deviations from normal dollar losses. This transformation requires interdisciplinary expertise to understand interactions between climate risk, risk perception and risk communication, but success ultimately hangs on sustained iteration of solutions between climate science and industry.

The Willis Research Network is engaged in an active research program to generate useful, useable and actionable seasonal climate information with the overarching goal to link public science with risk management to promote climate resiliency.

Although it will never be possible to predict the specific details of high-impact weather events such as Hurricane Sandy in advance of a season, there exists a wealth of untapped predictive information on shifts in their likelihood that presents tangible business opportunities.

Hurricane Sandy cumulative wind history courtesy of NOAA/NHC