Nowcasting: advances and applications

A Royal Meteorological Society National Meeting focusing on nowcasting advances and applications was held at the University of Leeds on 28 February 2023, hosted by Leeds Nowcasting Collective. The event, attended by academics and professionals, reviewed some of the challenges with current nowcasting practices and research. The meeting also highlighted some opportunities to advance work in this area. The meeting was opened by Leeds Nowcasting Collective coordinator Joseph Smith (University of Leeds). A photograph of meeting attendees is shown in Figure 1. Estelle De Coning (World Meteorological Organization) introduced the current state of nowcasting, illustrating how advancements in technology over time have led to big changes in this field of work. This included discussion of the increase in data availability, particularly how EUMETSAT's new geostationary satellite1 would increase the frequency of data – especially over Africa where radar data are sparse. As head of the World Weather Research Programme (WWRP)2 at WMO, Estelle presented some of the current nowcasting challenges faced worldwide. A key barrier is the limited capacity for nowcasting. Despite the advancement in satellites, there are problems with access to the data, Internet connectivity and training requirements. There was an interesting discussion on the increasing need for collaboration with private corporations for additional computing resources. The importance of working with these companies to advance nowcasting and how to go about this in the right way was stressed. It was noted that even in locations where access is not an issue, challenges remain with the dissemination of information: Were the warnings reaching the right people and the impacts being communicated effectively? The concluding remarks included a discussion on how the WWRP is working towards seamless prediction with engagement from endusers as an essential part of this. Seonaid Anderson (UKCEH) broke down the complicated nowcasting process into three simple questions: (1) Is there anything going on? (2) Is it going to get me? and (3) Is that going to be a problem? The first question addressed how we observe the current state of the weather. To forecast the weather effectively, having accurate observations is a crucial starting point. A whole range of products were introduced which use different methods to do this, some focusing on storm development and others on rainfall rates. Here, she mentioned the FASTA app3 which uses nowcasting to provide real-time information in East Africa. The second question explored how we predict which areas are going to be affected. Different methods have been developed to predict how the weather will develop including forward propagation through optical flow, statistical methods using the full climatology of the region and artificial intelligence (AI) methods. The final question addressed the impacts of the weather on different regions, and how we send out warnings. The use of testbeds to evaluate warning messages was highlighted as important for improving how warnings are sent. Seonaid emphasised the work UKCEH have been doing including their nowcasting portal for Africa4 which brings together some of the tools mentioned in one place. Following on from Seonaid, Claire Bartholomew (Met Office) explained in more depth how machine learning (ML) has played a growing role in nowcasting. In recent years, the use of ML models in nowcasting applications has increased with expansions in data set size, advancements in architecture complexity and diversification in data. The presentation showed how ML has been used in a range of applications including precipitation nowcasting, but also for tornado, lightning and fog prediction. Claire discussed an increase in models designed to improve the initial errors, including blurring precipitation fields at longer lead times and identification of extremes, but emphasised improvements can still be made. She stressed there was still a lot of opportunity for ML to be used in nowcasting. Some examples included building on existing traditional models that already show good predictive skill, the classification of storms (such as prediction of severity or duration) and using ML for hazard to impact translation. The talk highlighted how the use of ML has driven nowcasting in recent years and reiterated the importance of ML in advancing the field further. Katie Norman (Met Office) provided an overview of the rapidly evolving nowcasting developments that have come out of the Met Office over the last 5 years. The Met Office Nowcasting Project began in 2018 with a key aim to improve its services for predicting summertime convection with at lead time of at least 1–2h. First, the Cell Tracker tool was presented, which uses 3D reflectivity composites from the UK meteorological radar network. The Cell Tracker derives areas of high liquid water content (i.e. the core of a convective storm), tracks them every 5min and produces probabilistic nowcasts out to an hour ahead. Next, the mesoanalysis tool was presented, which combines a first-guess forecast with observations from across the UK, to produce estimates of temperature, dewpoint, pressure, wind and visibility. A unique feature of this tool is that many of the observations are crowd-sourced from members of the public through the Weather Observations Website.5 Finally, Katie discussed the importance of impact nowcasting and detailed some of the ongoing work into exploring localised emergency rainfall alerts. The Met Office nowcasting testbed, which is planned for the summer of 2024, will be a great opportunity to trial these new tools and make further improvements. The final presentation from Neil Ryan (Environment Agency) further elaborated on the challenges of providing warnings using nowcasts. The flood forecasting and warning service provides alerts for flooding from the main rivers and the sea, aiming for as long a lead time as possible whilst minimising false alarms. Neil discussed the challenges that have come with continued expectation on the service to provide warnings for smaller and smaller areas. Firstly, he stressed the importance of quick and effective communication. He highlighted that in a changing climate with an increasing number of unprecedented events, it was crucial that the public understand the warnings, impacts and when to act. Secondly, Neil discussed the difficulty in providing accurate warnings in situations with more dynamic weather and rapidly changing flood risk. The service currently uses nowcasts alongside observations and NWP. Neil questioned how nowcasts could better be used in their current framework and whether improving them would actually provide solutions for challenges in flood forecasting, setting up some interesting discussion. The presentation showed that the current methods for nowcasting are not effective in some applications and set challenging questions for the research community on how to address this. The meeting ended with a panel discussion with the invited speakers. Across this session, and the entire day, one common theme was the need for integrated nowcasting services which were co-designed with physical scientists, social scientists and end users. Improving the scientific aspect and accessibility of services for forecasters holds equal importance to the effective communication of warnings. It was emphasised that the nowcasting procedure is not linear but more cyclic with frequent feedback required and thus the ability to reshape the science based on the user needs. Interestingly, the meeting also brought attention to the fact that nowcasting is not necessarily the most useful tool for all applications, and what we can do to improve this is still an open question. The meeting marked an important step in bringing together researchers in nowcasting, highlighting the range of work happening and identifying directions for the future of this field. Kasia Nowakowska: Writing – original draft; writing – review and editing. Joseph Smith: Writing – original draft; writing – review and editing. There are no data sets in this article.