We’ve come a long way in weather forecasting since the 1850s, when Admiral Fitzroy, driven by the dangers he knew first-hand from his years as a sea Captain, used his experience and training to develop early weather stations and storm warnings for the maritime community. He saved lives and was widely regarded to have invented modern-day weather forecasts. We’ve gradually developed techniques and improved accuracy, buoyed by the advent of satellite and radar technology and super computing capability. Now we can not only forecast the weather out to four days as accurately as we could to one day 30 years ago1, but we can also predict the seasons months in advance. We can even use climate models project decades ahead into the future, to better understand the impacts of climate change
An international initiative2 on climate prediction led by the World Meteorological Organisation (WMO) and the Met Office in the UK, is driving the development of new services to industry to leverage the emergent skill in seasonal to decadal predictions. One of the new products, released just last week, is the “Global Annual to Decadal Climate Update” which looks five years into the future. This latest update describes the likely state of the climate until the end of 2024, and some of the predictions are put into the context of climate change though referencing the international targets agreed in Paris at the COP213, namely to keep global temperatures to less than 1.5 degrees Celsius above pre-industrial levels (defined by the global temperature average between 1850 and 1900) by cutting CO2 and greenhouse gas emissions.
The prediction is available via the WMO4 and reports a number of key global and regional findings. Globally the prediction says that in the next five years:
Amongst the detail of the report, it also confirms that we are currently globally around 1 degree Celsius warmer than pre-industrial times, and that it remains likely that the global temperature will remain at least 1 degree warmer in each of the next five years.
The five-year climate forecast brings together multiple forecast models from the UK, Spain, Germany, Canada, China, USA, Japan, Australia, Sweden and Denmark to provide a consensus view which reduces the impact of any biases present in individual models. Dr Leon Hermanson6 (former Willis Research Network fellow from 2011 through 2013) worked on this project and explains7 how the combination of all of these modelling centres from around the world allows the creation of a product that beats the output from any individual source. Dr Hermanson worked with the Willis Research Network (WRN)8 on decadal predictions, helping us to understand the predictability of the North Atlantic Oscillation9 and how we might use it to understand storm frequency ahead of the winter storm season in Europe. This work was part of Professor Adam Scaife’s10 team at the Met Office which focusses on monthly to decadal predictions, and the WRN continues to work Professor Scaife on a variety of projects relating to climate variability and seasonal predictions of storminess. The WRN has long supported new developments and applications of climate science to the finance sector, and with our academic partners, currently investigates the relationships between seasonal variability and observed insurance losses, and therefore whether new climate prediction techniques can provide insights useful in writing reinsurance contracts in terms of premium pricing or reinstatement options on annual and multi-annual timescales.
Forecasting of day-to-day weather relies on a mixture of weather and climate modelling techniques. For fine detail, high resolution numerical weather prediction (NWP) models are used to step forward in time from a well-defined initial condition as represented by the global weather observing network(weather stations, satellite and radar sensing, ocean buoys, weather balloons, and aircraft and ship observations etc). However, this representation isn’t perfect and observations can be biased, temporally inconsistent and spatially sparse (especially in regions with lower levels of investment in meteorological infrastructure). Therefore, the NWP models, are complemented by probabilistic methods where running models at lower resolution frees up computer resources to allow multiple iterations with slightly different initial conditions, designed to account for the uncertainty from a lack of precision in the observations. This creates an “ensemble” prediction and produces a probabilistic output, giving us a percentage chance of different phenomena to occur.
As we move from forecasting weather over days and weeks, to looking months and years ahead, the deterministic forecasting approach of a single high-resolution run has less value due to non-linear interactions in the atmosphere, essentially, chaos in the weather system. This creates a limit to how far into the future it is valid to forecast specific details of the weather11 of about two weeks, depending on the conditions. Therefore, probabilistic ensemble methods not only help with day-to-day forecasting, but are the main tool we have longer time scales, as the weather “noise” is smoothed and we can see seasonal, annual or even decadal variations, which can be predictable. Much of this predictability comes from the temperature inertia in the ocean surface layers, a type of memory that persists on longer timescales than the atmosphere. That said, atmospheric variations in variables such as tropical rainfall or meanderings of the jet streams high above us, can also give us information and shift the balance of probabilities to enable long range predictions. This is why seasonal forecasts use language that describes likelihoods for example, “likely” means ~70% chance, while “unlikely” is around ~20% probability.
The levels of skill are determined by applying the seasonal forecasting methodology to past seasons and comparing the results to what was observed. This is known as “hindcasting”, and is essential to understand the reliability of predictions and forecasts alike, and is similar to how daily weather forecasts are routinely verified against observed weather to help improve forecasting models and effective communication of the uncertainties. Decadal prediction science has been showing increased skill for some time, and it is now showing a reliable level of guidance12 and so it can start to become useful for decision makers in business and government policy, to aid in their strategic planning in weather and climate related risks.
The predictions discussed above are not the same as climate change projections made by the Intergovernmental Panel on Climate Change (IPCC)13 for their Representative Concentration Pathways14. Predictions are made using known parameters within which we can step forward weather and climate prediction models, bounded by conditions that we know will be virtually certain (accepting observational biases). Climate change projections are made using pretty much the same physical models, just at lower resolution still, again to save computing power to use on stepping further forwards, decades in to the future. The key difference is that on longer timescales, more uncertainties are introduced in terms of how society will evolve, and so projections need to change those parameters through time.
In the case of climate change projections, greenhouse gases are the key variable that may follow a number of different paths. The biggest uncertainties in determining how warm the world will become are derived from choices that human society makes with regards to reducing the burning of fossil fuels. So, different narratives are created ranging from a quick and orderly transition to a zero-carbon economy, all the way through to the continuation and expansion of fossil fuel extraction and combustion. These narratives are developed using socio-economic models known as Integrated Assessment Models which can relate global economic, political and social activity to the amount of energy we use and how much CO2 we burn to create that energy. Some are simple and some are hugely complex, but they all form the narrative to create a CO2 concentration pathway that we can then run through our physical models to represent how the climate will physically change. In this sense, they are not predicting the future, but instead projection a view of what the future might realistically look like, given the different choices societies will make.
The five-year predictions, described above, do not go far enough into the future to get to the point where CO2 pathways diverge according to how well humanity does at reducing greenhouse gas emissions, but they still include the influences of expected increases in atmospheric CO2 concentrations during the forecast period. CO2 has such a long lifetime in the atmosphere that the fossil fuels we burn now will have a lasting impact well beyond the five-year time horizon, so it is safe to assume continued increases for this period, while the world embarks on the transition to a low carbon economy. To illustrate this point, the impact of the economic slowdown due to Coronavirus on CO2 concentrations is only significant in the short-term “rate” of emissions15. The long residence time of CO2 in the atmosphere means that the concentrations of CO2 which cause climate change, will continue to rise as carbon emissions accumulate, even with temporary reductions in emission rates.
While academia advances the use of climate projections and develops ever more sophisticated climate scenarios, it is important to not confuse this approach with another methodology that is becoming popular in the finance sector. A “scenario approach” in finance circles describes the process of applying adjustments to current risk assessments, to represent climate change in the future and then use those scenarios to stress test asset portfolios. Stress testing is nothing new to the finance sector and every insurance company needs to have a view on the impact of a 1-in-200 year loss event for Solvency II16, for example. Stress testing is now being used to understand potential impacts of climate change on extreme events and consequently on insurance portfolios. For example, “Climate Change Scenarios” have been introduced through the General Insurance Stress Tests issued by the Prudential Regulation Authority (PRA)17. This is one of a number of steps towards embedding climate scenario-based risk assessment into strategic business planning to manage climate risks across the finance sector, as outlined in the recent letter to CEOs from the PRA18. The insurance industry now has guidance on how to assess the risks related to climate change without having to delve into the realms of climate science. As these financial stress test scenarios develop, they will need to become more aligned with the IPCC scenarios to enable the narratives behind the different projections to be relevant for strategic planning to manage climate risks.
In the corporate sector, the Task Force on Climate-related Financial Disclosures19 is also driving the transformation of shareholder reporting. Climate change is often dropped in to the Environmental bucket of Environment, Social and Governance (ESG) but it truly spans the E, the S, and the G which puts it squarely on the agenda of every company’s C-suite20. Where companies are not already feeling the effects of increased climate related loss, developing a climate resilient business strategy, or responding to public sentiment, then regulatory requirements will be a significant driver for action on climate change.
Willis Towers Watson has been advising its clients on how to respond to the regulatory requirements and providing consultancy services through the suite of capabilities under Climate Quantified21. The Willis Research Network underpins the development of climate risk analytical services and products, and helps with direct client interactions when needed, providing a seamless combination of tailored products and academic insight for each company that we support. While some aspects of climate risk can be generalised and used to diagnose broad areas of higher risk, detailed analysis is often bespoke and requires a nuanced approach using subject matter experts and new science combined with the Willis Towers Watson expertise in risk management and financial risk transfer.
Geoff joined Willis Towers Watson in 2013, and works with the Willis Research Network stakeholders and academic partners to match business needs to the latest in scientific research, and derive tangible outputs for Willis Towers Watson to help advise its clients to advance their understanding of risk from weather and climate related hazards.
His background is in meteorology and climate science, having worked in forecasting for over a decade for the UK Met Office and Bermuda Weather Service, in all aspects of delivering forecast services from media broadcasting to delivering warnings and actionable guidance on extreme weather phenomena such as tropical cyclones and heavy rainfall leading to flooding.
He holds a BSc in Environmental Science from the University of East Anglia, and a Masters (with distinction) in Climate Change from University College London. He is also an active Fellow of the Royal Meteorological Society.