Before looking for the drivers that might influence weather forecasting in the future, I thought it instructive to explore the past. In doing so, I am struck by the enormous changes that have occurred. Advances in science and technology have driven huge progress in accuracy, but changes in society have also altered the relationship between forecaster and user, while climate change, population growth and increasing wealth have increased the weather’s impact and hence the potential value of forecasts to society. When looked at closely, the primary drivers of these changes all appear to be technological, most notably computers, remote sensing and mobile phones. Some have impacted directly on the forecasting process, but others have had an indirect influence through society and its consumption of weather information. These developments have generally taken at least 25 years to reach maturity. Thus, the main drivers for the next 25 years should already be in place. It is just a matter of identifying which ones will turn out to be significant.

AI and the Future Forecast

Current developments in the application of Artificial Intelligence (AI) are evidently going to have a direct impact on the forecasting process. Machine Learning (ML) emulators have achieved a capability comparable to that of Numerical Weather Prediction (NWP) 60 years ago, in the sense that they outperform prior operational forecasting methods in some, but not all, respects. Over the next 25 years, I expect ML emulators to replace NWP in routine operational weather forecasting. Whether ML forecasting models will operate directly from observations or need separate data assimilation scheme remains an open question, but undoubtedly an AI-based data assimilation capability will be developed.

The migration from NWP to ML will challenge many currently accepted norms. It is conceivable that accuracy may trump consistency or conservation – at least for some customers – potentially resulting in a move away from seamlessness. AI may also more easily enable direct prediction of weather-related hazards, such as flood inundation, or weather impacts such as heat-related morbidity and mortality. While moves in this direction have been underway using impact models coupled to NWP, the switch to ML could result in the weather forecast becoming a hidden part of a broader customer-oriented impact forecasting system. The challenge will be in creating the training data, an area where I expect physical models to continue to have an important role.

Beyond the Forecast: Urban, Impact & Personalised Systems

With the move to ML in operational forecasting, there will be new opportunities for using the available computing power. Current experiments in urban prediction with hectometric NWP models may be a pointer, especially if a clear understanding of the requirements and benefits of intra-urban forecasts can be formulated. Alternatively, forecast skill may be advanced using very large ensembles. Or the focus may be on adding complexity by predicting weather-related hazards and their impacts. Some of these choices will be driven by changes in society’s consumption of weather information. Personally, I would like to see effort focussed on accurate forecasting of cloud-scale detail in the first few hours.

I have emphasised the use of ML in operational forecasting, but forecasts need to be validated and interpreted, and this requires an understanding both of the real meteorology and of forecast characteristics. A key challenge with a learning algorithm is that its performance may not be consistent enough for human quality control. However, there is increasing evidence that the performance of ML emulators can be made as stable as NWP models, enabling their characteristics to be learned in the same way as has been done for NWP models. The situation with respect to understanding the real world is more challenging. Through the use of model hierarchies built on exact solutions to the hydrodynamic equations, NWP has contributed greatly to our understanding of atmospheric behaviour, especially in the mid-latitude storm belts, but increasingly also in the Tropics. However, our understanding of many smaller scale processes, such as the growth and decay of convection or the formation and clearance of fog, remains poor. Progress may be achieved by continuing to increase the resolution and complexity of our research NWP models. However, in other spheres of science, knowledge is gained through rigorous use of statistical analysis, so it is my hope that AI methods may contribute to cracking some of these difficult problems.

AI will undoubtedly have a wider impact than just the forecast. The use of ML in retrievals of remotely sensed observations is a live area that promises substantially greater use of the information from hyper-spectral instruments. Indeed, AI offers many opportunities for exploitation of the new generation of Geostationary and Polar Orbiting satellites exemplified by Meteosat Third Generation and Metop2. AI also has the potential to facilitate the integration of a wider range of observations, including observations from drones (both in the atmosphere and the ocean) and the long awaited ubiquitous observing capability potentially offered by Internet of Things technology.

The delivery end of the forecasting and warning chain is ripe for more radical change. A key area of recent development has been the shift towards providing impact information. The benefit of impact- based warnings, over hazard threshold-based warnings, is that each person’s risk threshold is different and varies through time. However, achieving full benefit depends on forecasts and warnings being tailored to individual vulnerabilities. There have been some experiments in allowing users to tailor their own warnings by setting personal thresholds in an app, but this requires the user to understand their own vulnerability. Over the next 25 years, delivery of weather information will make use of Large Language Models such as ChatGPT, built into mobile phone apps and hubs such as Alexa. These will be configured to learn the behaviours of their users and to tailor the presentation of weather information accordingly. I doubt that this will be adopted for official warnings, due to the challenge of quality control, but the technology is available and, in time, many people will come to rely on it.

A Connected World: Climate, Energy & Autonomous Systems

The next technological development I will address is robotics, or more specifically, autonomous vehicles. Some of the wilder predictions of progress have been seriously knocked back, but there is so much investment in this area that it seems inevitable that on a 25-year timescale it will have an impact on many aspects of private and commercial transport. This is important because the sensor systems at the heart of autonomous transport are highly weather dependent. Some of this dependence will be removed through improved sensing technology and fail-safe systems, but an increased role for weather forecasts seems likely, with information going directly into vehicle control systems. This may be through a closed system of collecting and synthesising vehicle information, then feeding it back, without involvement of the weather service, but there would be clear benefits if vehicle observations could be used in the forecast and forecast information could be integrated into the vehicle control system. The benefits may be even greater for projected services using air- or ocean-borne drones, as it will almost certainly be necessary to put their operation into a safe mode in adverse conditions such as high winds or heavy rain.

Climate Change is important for weather forecasting for several reasons, not least the increasing frequency of hazardous weather events. However, I suggest that it is the technology of energy production and consumption, as we move towards a Net Zero world, that will have the greater impact on weather forecasting. Both solar and wind power are highly weather dependent. Reliable power supply requires forecasts of production a day in advance, and these are sensitive functions of weather variables. Power demand is also weather-dependent, especially in the variable weather of the UK, where winter heating requirements are very temperature sensitive, and summer cooling requirements may become more so.

Risks, Resilience & the Role of Meteorologists

These technology driven developments have a lot of momentum behind them, but their widespread adoption is dependent on public investment in supporting infrastructure. Delivery of the required information also depends on continued investment in weather services. So, a potential disrupter of these developments is a crisis in government funding, such as might occur due to a collapse of international trade, a war involving the UK, or sustained loss of business confidence. Currently the world seems to be moving into a state that makes one or more of these more likely. On the other hand, if security tensions increase, the military applications of these technologies might lead to accelerated demand for relevant weather information.

While the technology provides the opportunities for these developments, it also creates risks which might hinder their implementation. An increasing proportion of compute power is already spent on neutralising viruses and repelling hackers, while much communication effort is now directed to correcting misinformation. Looking forward, such threats could pose a serious challenge to the safety of autonomous vehicles, while massive financial fraud could lead to business collapse, and misinformation could result in civil unrest or war.

Despite a general progression towards greater insulation of human life from the weather, public surveys have shown a marked increase in the use of and dependence on weather information. Partly, this can be attributed to a growing trust in the available information, which enables people to use forecasts with greater confidence. However, there are several other influences which might be expected to continue this trend into the future. As the global population becomes more urbanised and increasingly dependent on infrastructure services, exposure to the weather is becoming less visible, and vulnerability is dependent on a wider range of hazards and combinations of hazards, e.g. the potential impacts of space weather on communication and power services, or of remote wildfires on water supply. At the same time, there is a declining acceptance of risk, both personally and corporately, often manifested in the desire to attribute blame and, with an ageing population, health issues are growing in importance. Many aspects of both physical and mental health are weather-related and there is scope for moving beyond warnings of hot and cold waves to provide much greater personalised information that enables people both to minimise the risks of disease and achieve greater wellness.

These drivers will also increase the movement towards a mixed economy of weather information. While the provision of safety of life advice will remain a public good, provided equally to everyone as part of the government commitment to the security of its citizens and paid for through taxes, commercial services will grow in number and scope, accelerated by applications of AI. A key issue for these services will be building and maintaining trust. Broadening of the scope of the weather enterprise across weather-related impact and human response, will demand a strengthened contribution of the social sciences, especially experts from the behavioural, economic and health sciences, and offer important roles for the arts and humanities in an increasingly complex web of endeavour, that should aim to equitably support the lives of all citizens,

To finish, I would like to ask my crystal ball how working in weather forecasting is likely to change. There has been rapid change in recent years, firstly from forecast automation, and then from remote working during and after the COVID pandemic. As AI extends the range of automation, I expect the role of the operational meteorologist will continue to evolve towards user outcomes. While this will require new skills in the effects of the weather on society and on how individuals respond to forecasts, it will still be critical for the meteorologist to understand the physical context of the predicted weather. Maintenance of this broader range of skills will require increased investment in continuous professional development. In 25 years’ time, the meteorology profession, will attract a wide range of talented individuals who feel called to contribute their particular skills to helping people be prepared for whatever the weather will bring, and the best meteorologists will combine their scientific expertise with the ability to create a compelling story that is memorable and leads to effective action.

Professor Brian Golding OBE FRMetS, President of the Royal Meteorological Society