Andrew Paul Barnes

Comparisons between observed and model‐resolved gravity waves (GWs) are crucial for evaluating general circulation model (GCM) simulation accuracy and understanding wave characteristics. However, observational noise often obscures waves, complicating such comparisons. To address this, we have developed a GW detection method using a convolutional neural network (CNN). The CNN is trained on Atmospheric Infrared Sounder (AIRS) temperatures with labels indicating wave presence based on Berthelemy et al. (2025, https://doi.org/10.5194/egusphere‐2025‐455 ). Their method detects noise‐induced pixel‐to‐pixel variations in horizontal wavelengths; in contrast, the CNN robustly identify waves even when applied to smoothly varying model data. Using this method, we compare stratospheric GWs in boreal winters between AIRS observations and a high‐top GW‐permitting GCM, Japanese Atmospheric GCM for Upper Atmosphere Research (JAGUAR). The results agree well and exhibit similar interannual variability, with discrepancies also identified, including a more zonally elongated distribution of tropical GWs in JAGUAR. This method is broadly applicable to the future use of satellites for guiding wave‐resolving atmospheric model development.

This study evaluates a new approach to improving streamflow forecasting with deep learning: it focuses on the novel application of a dynamic multimodal feature fusion mechanism that adapts fusion operations based on the data's characteristics. Two baseline Long Short-Term Memory (LSTM) architectures are used, applying two dynamic fusion methods: dynamic operation-level fusion and attention-based fusion, to combine heterogeneous and multisource hydrometeorological data. The models are used for univariate (single flow gauge) and multivariate (multi-gauge) streamflow forecasting approaches. Applying these four approaches to the Severn Basin in the UK, known for long medium- to high-flow periods and shorter low-flow intervals, shows that the dynamic operation-level fusion consistently improved over the attention-based fusion in key performance metrics. In the multivariate case, Nash-Sutcliffe Efficiency (NSE) improved by 1.43%, Mean Absolute Error (MAE) decreased by 1.73%, Mean Absolute Scaled Error (MASE) dropped by 1.82%, and high-peak MAE decreased by 3.36%. For the univariate case, NSE improved by 1.44%, MAE decreased by 4.02%, MASE dropped by 3.89%, and high-peak MAE improved by 2.8%. In addition, multivariate models were considerably faster than univariate models, with training and inference times reduced by 74.57% and 73.81%, respectively. The multivariate models showed a 2.75% increase in NSE and a 72.04% decrease in MASE, indicating they captured better the hydrologic variability than the univariate models. Conversely, univariate models had a 20.59% lower MAE, a 21.17% lower high-peak MAE, and greater stability as indicated by tighter interquartile ranges, suggesting better error minimisation and more reliable predictions. Notably, in two river stations all models underperformed due to rapid flow variability and flashy hydrological responses in smaller catchment areas, suggesting in the future the use of higher-resolution climatic data. Overall, the study shows the potential of new dynamic multimodal fusion techniques, navigating the operational trade-offs between speed, stability, and accuracy across multi and uni-variate training strategies in streamflow forecasting. Nonetheless, the need for an optimal operational balance remains, suggesting further refinement of fusion techniques and focusing on minimising uncertainty.

Climate change is changing rainfall and flood regimes across the world with severe and widespread impacts on society. Rainfall extremes are intensifying in frequency and magnitude due to the effects of climate change, and thus in this research, we introduce a new, novel framework for understanding how rainfall distributions are changing through time, enabling more accurate flood risk analysis. The framework offers two approaches to comparing rainfall distributions, the first of these utilises a stagnant benchmark distribution and the second highlights a moving benchmark approach. When combined the framework enables the identification of significant sudden and gradual changes in the distributions without the need to fit statistical distributions to the data. The region of Europe is selected as the case study and analysed in the four UN regions of Europe: Northern Europe, Eastern Europe, Southern Europe, and Western Europe. Using daily precipitation data generated using the ERA5 Reanalysis hourly data from the ECMWF’s Copernicus data store, the case study is used to highlight the capability of both frameworks to capture different forms of rainfall distribution shift. Comparing the frameworks presented revealed similar long term changes in the rainfall variation. The stagnant comparison showed that rainfall distributions have intensified since 1940 with a clear increase across all four regions of Europe regarding the percentage of days with rainfall, averaging at 2.75% across Europe. The largest changes seen are in the last comparison period for Eastern Europe (1960-1975) at 3.07% and in the latest comparison period (2005-2020) for Northern Europe (2.64%). The moving comparison method unveiled the strongest changes between the periods 1940-1960 and 1960-1980 with an average of 2.09% of rainfall days being intensified across all Europe. The most considerable shifts in rainfall variability occurred in Eastern (2.39%) and Western Europe (2.72%) during the 1960-1980 period. By applying it over the European region, this paper demonstrated how this novel approach can be used to identify long-term rainfall variation in the 20th century. The suggested frameworks do not rely on fitting statistical distributions and thus enable both long and short term change identification, providing flood risk managers a new solution to understanding local, regional and global rainfall variability and quantification. The analysis of the changing dynamics of precipitation patterns and the increase of the intensity of precipitation events, offers considerable potential for further investigations in the mitigation strategies of a resilient future.

This study introduces image‐based classification techniques to identify whether trash screens in urban rivers are blocked. The study obtained 755 images from a CCTV camera surveying a trash screen located on an urban river at Tongwynlais in Cardiff. Manual quality control reduced the dataset to 577 images, labelled as either blocked (80%) or unblocked (20%). The performance of a logistic regression for classification of images was investigated using three different subsets of the labelled images: (1) the original dataset, (2) a balanced but under‐sampled dataset with equal number of blocked and unblocked images, and (3) an augmented dataset with an equal number of blocked and unblocked images using Gaussian noise augmentation to increase the number of unblocked images. Results show that our data‐augmentation method enhanced model accuracy by 8%, successfully classifying images as blocked or unblocked with an accuracy of 88%; by overcoming the bias in the dataset these results also highlight potential solutions to overcome the challenges of operating this methodology across a network of cameras. This enables authorities in both data rich and data scarce regions the ability to take advantage of machine learning to open up the next generation of a distributed, data‐driven flood warning systems, protecting people, infrastructure and the environment.

Anomaly detection is used to explore the link between data-driven anomalous storms and their socio-economic impact on countries within the North-West Pacific. Three anomaly detection models are trialled using three distinct algorithms on the storm tracks and temperature profiles of storms. A feature-based comparison of the top 5% of anomalous storms from each model is used to reveal variations in anomalous storm activity. Further to this, the socio-economic impact of the anomalous storms is assessed, revealing a link between the anomalous behaviour of storms and the impact experienced by countries on their path. A final cross-comparison shows that the k-Nearest Neighbour and Isolation Forest algorithms succeeded at identifying high-impacting storms. However, the agglomerative clustering model found many unique storms that had low impact. This highlights the importance of considering both trajectory and temperature in determining the severity and impact of erroneous storms.

Traditional weather forecasting approaches use various numerical simulations and empirical models to produce a gridded estimate of rainfall, often spanning multiple regions but struggling to capture extreme events. The approach presented here combines modern meteorological forecasts from the ECMWF SEAS5 seasonal forecasts with convolutional neural networks (CNNs) to improve the forecasting of total monthly regional rainfall across Great Britain. The CNN is trained using mean sea-level pressure and 2-m air temperature forecasts from the ECMWF C3S service using three lead-times: 1 month, 3 months and 6 months. The training is supervised using the equivalent benchmark rainfall data provided by the CEH-GEAR (Centre for Ecology and Hydrology, gridded estimates of areal rainfall). Comparing the CNN to the ECMWF predictions shows the CNN out-performs the ECMWF across all three lead times. This is done using an unseen validation dataset and based on the root mean square error (RMSE) between the predicted rainfall values for each region and benchmark values from the CEH-GEAR dataset. The largest improvement is at a 1-month lead time where the CNN model scores a RMSE 6.89 mm lower than the ECMWF. However, these differences are exacerbated at the extremes with the CNN producing, at a 1-month lead time, RMSEs which are 28.19 mm lower than the corresponding predictions from the ECMWF. Following this, a sensitivity analysis shows the CNN model predicts increased rainfall values in the presence of a low sea-level pressure anomaly around Iceland, followed by a high sea-level pressure anomaly south of Greenland.

This study presents a new methodology for improving forecasts of current monthly, regional precipitation using video-based convolutional neural networks (CNNs). Using 13 administrative regions of Great Britain as a case study, three CNN architectures are trained for each region to forecast monthly rainfall totals given forecast mean sea-level pressure and 2-m air temperature videos from the MetOffice GloSEA5 model and a benchmark rainfall data. The forecasts generated by the CNN and the GloSEA5 precipitation forecasts are both compared directly against a benchmark rainfall dataset for each of the regions. Following this, the CNN models are combined with the GloSEA5 forecasts to generate a new ensemble for each region which is then compared with the benchmark rainfall. The results show that the trained CNNs produce errors similar to the GloSEA5 model with RMSEs of 63 mm (single frame), 44 mm (slow fusion), and 37 mm (early fusion) compared with the GloSEA5 error of 33 mm. Regional variability remained consistent throughout the compared models. However, the CNN models all outperform GloSEA5 in the prediction of extreme events. Furthermore, treating the forecasts as an ensemble results in errors of 32 mm (CNN ensemble) and 31 mm (post-processing ensemble), both of which improve on the independent GloSEA5 forecasts.

Severe flooding in the United Kingdom is often linked to the occurrence of heavy rainfall events, which can be characterized by the synoptic scale meteorological conditions over the North Atlantic region. Seasonal heavy rainfall events (summer and winter 1‐day maxima) were extracted from 125 locations across Great Britain over the period 1950–2017. For each event, anomaly sea‐level pressure and 2 m air temperature conditions across the North Atlantic sector were extracted. In contrast to earlier studies, these two datasets were combined and clustered to identify how the pressure and temperature conditions co‐vary within each half‐year to produce heavy rainfall events. Distinctly different spatial patterns were found for four classes in summer and for three classes in winter. For all classes there is a negative sea‐level pressure anomaly centred over or near the British Isles. However, whereas in summer the low pressures are associated with either predominantly cold or warm anomalies over most of the North Atlantic, in winter two phases of a smaller‐scale four‐pole temperature pattern emerges. Nevertheless, for one of the winter classes the cold anomaly over the northwest Atlantic is so deep, persistent and widespread that, unusually for the winter season, a significant relationship between the class's frequency of occurrence and the Atlantic Multidecadal Oscillation (AMO) index is found (r = −.39). Further, for both seasons heavy rainfall occurs when the AMO and the North Atlantic Oscillation (NAO) are in opposing phases. Particularly, positive NAO and negative AMO result in heavy rainfall in western Britain. Two classes in each season are consistent with positive and negative phases of the NAO, and the two non‐NAO summer classes are associated with a northward extension of the subtropical high pressure and heavy rainfall in the southeast. The association between heavy rainfall and large‐scale circulation and temperature drivers can find application in, for example, weather generators.

This study presents the first attempt to identify extreme rainfall events based on surrounding sea-level pressure anomalies, using neural network-based classification. Sensitivity analysis was also performed to identify the spatial importance of sea-level pressure anomalies. Three classification models were generated: the first classifies the patterns between extreme and regular rainfall events in the North West of England, the second classifies the patterns between extreme and regular rainfall events in the South East of England, and the third classifies between the patterns of extreme events in the North West and South East of England. All classifiers obtain accuracies between 60 and 65%, with precision and recall metrics showing that extreme events are easier to identify than regular events. Finally, a sensitivity analysis is performed to identify the spatial importance of the patterns across the North Atlantic, highlighting that for all three classifiers the local anomaly sea-level pressure patterns around the British Isles are key to determining the difference between extreme and regular rainfall events. In contrast, the pattern across the mid and western North Atlantic shows no contribution to the overall classifications.