Lesley De Cruz

The predictability of the generative AI-based nowcasting model LDCast is evaluated over Belgium, together with the pysteps implementation of the nowcasting algorithm STEPS. Neither STEPS nor LDCast were fine-tuned for the Belgian region, so both models are evaluated under conditions in which they will most likely be used in practice at national weather offices. STEPS and LDCast are slightly underdispersive, but the ensemble spread provides an estimation of the error at almost all scales. Both models adapt the properties of their ensembles to the type of event, either convective or stratiform. The spatial scores of the STEPS and LDCast ensembles are compared with those of surrogate ensembles, revealing that both STEPS and LDCast have very little ability to spatially localise the error of the ensemble mean. This suggests that the content of STEPS and LDCast ensembles is informative in terms of statistics, but not in terms of dynamics.

The rapid expansion of wind farm installations in the North Sea results in an increased need for understanding their influence on the local atmosphere, as well as the interactions between them. Wind farm operation and power production are affected by wakes produced both within and upstream of the wind farms. Accurately estimating these impacts requires robust mesoscale atmospheric modeling capable of capturing extended wind speed deficits and their dependence on atmospheric stability. This study presents a 3-year-long mesoscale analysis using the Weather Research and Forecasting (WRF) model at a horizontal resolution of 1 km, a comparatively long and high-resolution configuration within the context of offshore wake research. The simulations are evaluated against four lidars located in the Southern Bight of the North Sea in the vicinity of the 3.7 GW Belgian–Dutch offshore wind farm cluster, providing an extensive observational basis that covers upstream, intra-farm, and downstream flow conditions. Coupling the mesoscale atmospheric model with the Fitch wind farm parameterization scheme improves simulation accuracy, particularly in regions frequently affected by wake effects. An evaluation across different atmospheric boundary layer stability regimes shows that the model performs best under less extreme stability, while a detailed assessment of upstream, intra-cluster, and downstream wake characteristics highlights the added value of the Fitch scheme for multi-year offshore applications. Finally, synthetic aperture radar images from selected wake events are compared with the model output, demonstrating that the wind farm parameterization scheme effectively captures the larger-scale structure and wind speed deficit of the wind farm wake at the analyzed timestamps. Together, these results provide one comprehensive long-term evaluation of offshore wake behavior in a dense wind farm cluster, helping to address the current gap in mesoscale wake studies.

The predictability of the generative AI-based nowcasting model LDCast is evaluated over Belgium, together with the pysteps implementation of the nowcasting algorithm STEPS. Neither STEPS nor LDCast were fine-tuned for the Belgian region, so both models are evaluated under conditions in which they will most likely be used in practice at national weather offices. STEPS and LDCast are slightly underdispersive, but the ensemble spread provides an estimation of the error at almost all scales. Both models adapt the properties of their ensembles to the type of event, either convective or stratiform. The spatial scores of the STEPS and LDCast ensembles are compared with those of surrogate ensembles, revealing that both STEPS and LDCast have very little ability to spatially localise the error of the ensemble mean. This suggests that the content of STEPS and LDCast ensembles is informative in terms of statistics, but not in terms of dynamics.

The predictability of the generative AI-based nowcasting model LDCast is evaluated over Belgium, together with the pysteps implementation of the nowcasting algorithm STEPS. Neither STEPS nor LDCast were fine-tuned for the Belgian region, so both models are evaluated under conditions in which they will most likely be used in practice at national weather offices. STEPS and LDCast are slightly underdispersive, but the ensemble spread provides an estimation of the error at almost all scales. Both models adapt the properties of their ensembles to the type of event, either convective or stratiform. The spatial scores of the STEPS and LDCast ensembles are compared with those of surrogate ensembles, revealing that both STEPS and LDCast have very little ability to spatially localise the error of the ensemble mean. This suggests that the content of STEPS and LDCast ensembles is informative in terms of statistics, but not in terms of dynamics.

In recent years, wind energy has emerged as one of the most essential energy sources in the transition towards a low-carbon future for climate change mitigation. Both onshore and offshore wind farm installations have expanded rapidly over the past decade, and they are considered a crucial provider in a cleaner energy mix. Specifically, the offshore wind farm installations that are currently operational and those planned for the near future over the North Sea will result in dense structures with a large number of wind turbines. While many studies, using both measurements and mesoscale models, have shown that wind farm wakes can substantially influence neighboring farms by reducing wind speeds and enhancing turbulence, other flow phenomena also play an important role. Among these are atmospheric gravity waves (AGWs), which can modify the surrounding flow and impact turbine operation. Although the concept of wind‑farm‑induced AGWs has been explored over the past decade, most existing research has relied on idealized large‑eddy simulations or reduced‑order models. Consequently, fundamental questions remain regarding how wind farm-triggered AGWs form under realistic atmospheric conditions, how frequently they occur, and how strongly they influence wind‑farm performance.In this work, we use lidar observations from five offshore platforms and satellite aperture radar (SAR) data to assist in examining multi-year mesoscale simulations using the Weather Research and Forecasting (WRF) model at a spatial resolution of 1 km. WRF driven by ERA5 over a multi‑year period reproduces realistic atmospheric variability and stability regimes, thereby supporting the assessment of the conditions that help wind-farm-triggered AGWs to develop. Our study suggests the AGW signatures appear to be either triggered or intensified by the presence of a large offshore wind‑farm cluster. Building on these findings, we propose a framework to evaluate AGWs that originate from, or are strengthened by, the 4 GW Belgian–Dutch offshore wind‑farm cluster in the Southern Bight of the North Sea. Preliminary results show that, although WRF tends to misrepresent the amplitude of AGW-related perturbations relative to lidar measurements, the model reliably captures the spatial and temporal patterns of AGWs in several analyzed events. These events were identified based on AGW‑like structures visible in SAR imagery and in the year‑long WRF simulations, and subsequently validated against the available lidar datasets. To support this identification, we estimate the atmospheric stability conditions and AGW wavelengths, while additional diagnostic methods, such as power‑spectrum evaluation, coherence analysis, and autocorrelation, are applied at the lidar sites. These promising results demonstrate the potential of our approach for advancing the detection and understanding of AGWs in real offshore wind‑farm environments.

Machine learning (ML) and artificial intelligence (AI) have become integral to modern weather and climate science, and their rapid evolution is driving new opportunities for urban climate research. Across the urban climate community, ML methods of varying complexities are being used for tasks ranging from fast point-based predictions to neighborhood- or city‑scale urban climate modelling for one or multiple (bio)meteorological variables. For instance, novel ML and AI approaches support scenario generation for impact studies such as quantifying the effects of urban vegetation, assessing thermal comfort under different warming pathways, or identifying locations for future cool spaces. In addition, ML techniques are also being applied to provide boundary conditions for micro‑scale models. While the field is expanding quickly, it remains highly fragmented. AI/ML in urban climate research spans diverse methods, scales, datasets, and scientific aims, making it difficult to define a common research agenda or benchmarks. The lack of an established interdisciplinary network between AI and urban climate communities further increases the risk of duplicated efforts and missed opportunities for coordinated progress. The newly founded AI4UrbanClimate working group addresses this gap by establishing an international community focused on AI/ML applications in urban climate research. Our goals are to (1) bring people together with similar research interests, (2) develop a common understanding of the scope and landscape of ML-based urban climate research, (3) review existing work across domains, and (4) identify persistent challenges and priorities and initiate and coordinate collective action - such as benchmarking datasets, standardized evaluation metrics, and best‑practice guidelines. As an initial step, we are mapping ongoing activities across the community to enable the development of meaningful benchmarks and identify where joint efforts could accelerate research. We will present the first outcomes from the initiative’s activities, including insights on who is currently involved and thoughts that were shared during the kick‑off meeting and online social event. We will also outline how you can join the AI4UrbanClimate network and contribute to building this emerging community.

The predictability of the generative AI-based nowcasting model LDCast is evaluated over Belgium, together with the pysteps implementation of the nowcasting algorithm STEPS. Neither STEPS nor LDCast were fine-tuned for the Belgian region, so both models are evaluated under conditions in which they will most likely be used in practice at national weather offices. STEPS and LDCast are slightly underdispersive, but the ensemble spread provides an estimation of the error at almost all scales. Both models adapt the properties of their ensembles to the type of event, either convective or stratiform. The spatial scores of the STEPS and LDCast ensembles are compared with those of surrogate ensembles, revealing that both STEPS and LDCast have very little ability to spatially localise the error of the ensemble mean. This suggests that the content of STEPS and LDCast ensembles is informative in terms of statistics, but not in terms of dynamics.

The predictability of the generative AI-based nowcasting model LDCast is evaluated over Belgium, together with the pysteps implementation of the nowcasting algorithm STEPS. Neither STEPS nor LDCast were fine-tuned for the Belgian region, so both models are evaluated under conditions in which they will most likely be used in practice at national weather offices. STEPS and LDCast are slightly underdispersive, but the ensemble spread provides an estimation of the error at almost all scales. Both models adapt the properties of their ensembles to the type of event, either convective or stratiform. The spatial scores of the STEPS and LDCast ensembles are compared with those of surrogate ensembles, revealing that both STEPS and LDCast have very little ability to spatially localise the error of the ensemble mean. This suggests that the content of STEPS and LDCast ensembles is informative in terms of statistics, but not in terms of dynamics.

The predictability of the generative AI-based nowcasting model LDCast is evaluated over Belgium, together with the pysteps implementation of the nowcasting algorithm STEPS. Neither STEPS nor LDCast were fine-tuned for the Belgian region, so both models are evaluated under conditions in which they will most likely be used in practice at national weather offices. STEPS and LDCast are slightly underdispersive, but the ensemble spread provides an estimation of the error at almost all scales. Both models adapt the properties of their ensembles to the type of event, either convective or stratiform. The spatial scores of the STEPS and LDCast ensembles are compared with those of surrogate ensembles, revealing that both STEPS and LDCast have very little ability to spatially localise the error of the ensemble mean. This suggests that the content of STEPS and LDCast ensembles is informative in terms of statistics, but not in terms of dynamics.