Data modeling for lightning strike severity prediction for severe storm and early tornado warning

With the advancement of modern sensing technology and network communication, the geological disaster monitoring and early warning technology is changing toward intelligence, continuity, and real-time monitoring. The most effective way to improve real-time monitoring and early warning is to conduct front-end analysis in the field of geological disasters. With the application of the Internet of Things technology in geological disasters, field data collection and field-level data strategies belong to the edge computing gateway for execution. According to the latest research trend of landslide on-site monitoring and early warning, edge computing gateways usually use thresholds for monitoring and early warning. On the basis of the empirical model of simple statistics or the theoretical model of landslide deformation early warning, key early warning indicators and weights are proposed to judge the stage of landslide evolution or disaster level. This study starts with a comprehensive analysis of the temporal and spatial evolution of the surface deformation area in the field of geological disaster monitoring. Then, an edge computing smart gateway is designed on the basis of the Linux operating system. The Long Range Radio field wireless sensor network is used to develop an automated monitoring system for slopes of displacement monitoring points, rainfall, and soil moisture content-inducing factors. Subsequently, the effective methods of obtaining multilevel topographic data of slopes are investigated using field information, such as the velocity and acceleration of surface displacement data. The slope multielement trigger monitoring model with multiple induced element coupling modes is explored to realize the front-end edge computing strategy method of slope monitoring and early warning. The analysis results can be sent to high-level data analysis servers, such as the software deployed on the Local Area Network and cloud servers of the relevant department for secondary use. This research overcomes the problems of regular collection of existing on-site monitoring instruments, loss of catastrophic information, and low effective data rate. Compared with traditional working modes, such as uploading the original monitoring data to a server for data extraction and analysis and then feedback to control the field equipment, using front-end edge computing is more efficient in realizing the first data processing utilization and adaptive monitoring processing. It can also provide more effective and reliable data for disaster monitoring and early warning. In conclusion, this research is of great significance in effectively capturing the laws of early evolution, improving real-time early warning capabilities, and promoting the industrialization of national geological disaster monitoring and early warning technology and equipment.

Natural disasters, such as earthquakes, floods, wildfires, and hurricanes, are occurring more frequently and becoming more severe, underscoring the need for advanced real-time prediction systems. Existing static monitoring infrastructures with limited coverage suffer from delayed processing and expensive maintenance costs. In comparison, mobile sensor networks (MSNs) offer cost-effective, dynamic, and scalable solutions for collecting real-time environmental data. This research focuses on the design and implementation of mobile sensor networks for real-time disaster prediction, with a particular emphasis on advanced sensing technologies and predictive algorithms. The methodology involves the deployment of mobile sensor units equipped with environmental sensors, communication modules, and data processors to monitor temperature, humidity, seismic activity, and air quality. These data are analyzed in real-time with machine learning algorithms to detect early warning signs of potential disasters and are centralized for further processing. System evaluation is performed through simulating disaster scenarios and is benchmarked against conventional fixed-network methodologies. Results indicate marked improvements in prediction accuracy, response time, and adaptability to rapidly changing conditions. This work analyzes the practical applications of MSNs in disaster-prone areas, discusses challenges such as energy consumption and data protection, and provides suggestions for future improvements. This study extends the expanding area of intelligent disaster management and mobile sensor networks to enhance early warning and response systems.

Massive manycore embedded processors with network-on-chip (NoC) architectures are becoming common. These architectures provide higher processing capability due to an abundance of cores. They provide native core-to-core communication that can be exploited via message passing to provide system scalability. Despite these advantages, manycores pose predictability challenges that can affect both performance and real-time capabilities. In this work, we develop efficient and predictable group communication using message passing specifically designed for large core counts in 2D mesh NoC architectures. We have implemented the most commonly used collectives in such a way that they incur low latency and high timing predictability making them suitable for balanced parallelization of scalable high-performance and embedded/real-time systems alike. Experimental results on a single-die 64 core hardware platform show that our collectives can significantly reduce communication times by up to 95 % for single packet messages and up to 98 % for longer messages with superior performance for sometimes all message sizes and sometimes only small message sizes depending on the group primitive. In addition, our communication primitives have significantly lower variance than prior approaches, thereby providing more balanced parallel execution progress and better real-time predictability.

- Crime prediction has become a critical challenge as rising crime rates threaten community safety worldwide, with traditional manual pattern identification methods proving inadequate for modern law enforcement needs. This research presents a novel real-time crime prediction system that leverages unsupervised machine learning algorithms, specifically Apriori and Apriori TID, to automatically recognize hidden patterns and correlations between different crime types. The methodology involves preprocessing crime datasets from multiple sources, using mining techniques for association rules to extract frequent crime combinations, and implementing the system as a web-based application using Microsoft technologies. The system analyzes historical crime data to generate predictive insights about future criminal activities and their interconnected relationships in specific geographical locations. Key results demonstrate the system's ability to determine the strong correlations between crime types with high confidence levels, such as correlations between robbery, kidnapping, and pickpocketing in particular areas. The comparative analysis reveals superior accuracy and efficiency compared to existing static prediction models. In contrast to traditional methods that depend on manual analysis, this automated system provides law enforcement agencies with real-time predictive capabilities, enabling proactive crime prevention strategies. The implementation as a functional web application distinguishes this work from purely theoretical studies, offering practical deployment potential for police departments. Key Words: Data Science, Machine Learning, Crime Prediction, Association Rule Mining and the Apriori Algorithm, Unsupervised Learning.

Early warning systems (EWSs) are important for risk management. This chapter develops a framework of an ecotourism EWS which includes hazard risk, carrying capacity of service facilities and tourism sites, and tourism development impact assessment, monitoring and warning. Tourism impact is an emerging research spot. In EWSs, negative tourism impacts should be emphasized, including environmental, economic, and social impacts. After the framework is constructed and analyzed, cases based on local characteristics are selected and the concentrated points for ecotourism early warning are analyzed.

Nepal's diversified terrain and vast watershed create both opportunities and obstacles for water resource management. Conventional methods frequently fail owing to insufficient data, old methodologies, plus a lack of compatibility with modern technology. To solve these issues, the research suggests using improved hydrological simulation and sophisticated models for forecasting for sustainable water resource management in Nepal. The study uses cutting-edge technology, such as satellite imagery, Geographic Information Systems (GIS), with neural networks, to create a new combination CNN-LSTM technique. Hydrological statistics regarding precipitation, river flow, and glacial melt, as well as meteorological, geographical, and socioeconomic information, was gathered then filtered. This data integration enhances the accuracy of models, providing real-time monitoring and prediction capabilities. The hybrid CNN-LSTM system incorporates the qualities of Convolutional Neural Networks with collecting time properties from images, CatBoost for dealing with tabular information and spatial traits, with LSTM for enhanced classification of images and sequence data processing. The suggested model beats previous techniques, with an estimation accuracy of 99.20% to identify hydrologic occurrences. The incorporation of these latest innovations enhances floods projections, famine projections, and general handling of water techniques. This study shows that sophisticated hydrological simulations may considerably improve the durability and long-term viability of the management of water resources in Nepal, making it a valuable tool for regulators and managers. The research results call for bigger investments in the internet, training, and the creation of favorable regulations to assure the long-term sustainability of these programs. By tackling the constraints of existing approaches and adopting creative approaches, Nepal may accomplish responsible management of water resources while ensuring the availability of water and increasing the standard of life over its people.

<p>本研究以中央氣象署於預報作業使用之對流監測平台系統(System for Convection Analysis and Nowcasting, SCAN)提供的歷史對流胞及冰雹預警資訊為基礎，結合由民眾或新聞媒體所記錄下來的歷史冰雹目擊事件，針對臺灣近年夏季冰雹的即時預警狀況進行統整評估。分析資料方面，為能涵蓋臺灣南北地區之事件，故分別選用五分山雷達、七股雷達作為北部與南部地區的資料來源。統計結果顯示，臺灣夏季冰雹多發生於弱綜觀天氣條件的下午時段，且實際目擊的冰雹尺寸平均值小於現行預警門檻。冰雹相應對流胞的持續時間比一般夏季對流更長，其中又以較低海拔的七股雷達訊號更完整。然而，更高的雷達掃描時間解析度，使五分山雷達的預警表現明顯更好。整體而言，過去SCAN系統於臺灣冰雹事件發生前可提供的預警時間多在15分鐘以內，核心問題則包含(一)部分目擊事件未成功預警以及(二)現有系統假警報過多二者。本研究針對上述情況，分別提供(一)以冰雹目擊個案特徵統計結果下修預警門檻與(二)以臺灣探空資料修正SCAN系統背景預設之環境0℃、-20℃高度等改善建議。</p> <p> </p><p>This study is based on the historical storm cell and hail warning information provided by the System for Convection Analysis and Nowcasting (SCAN), which is utilized by the Central Weather Bureau (CWB) for operational forecasting. By combining hail records from recent years as reported in the news, the real-time early warning capabilities for Taiwan’s summer hail can be integrated and improved. In terms of the analysis data, Wu-Fan San (RCWF) and Chi-Gu (RCCG) were selected as radar data sources to cover events in the northern and southern regions, respectively.<br />The results indicate that most of the events occur in the afternoon during weak synoptic weather conditions, and the size of witnessed hail is mostly below the warning threshold. Hailstorms have a longer duration compared to general summer storm cells, and the RCCG at lower altitudes provides a more complete signal. However, the higher time resolution of the radar scan gives RCWF a significantly better early warning performance. In summary, SCAN can provide a lead time of approximately 15 minutes for hail warnings, but two major issues are identified: (1) some hail events were not detected by the SCAN system, and (2) there are too many false alarms. This research proposes two solutions: (1) lowering the warning threshold based on the statistical analysis of hail reports, and (2) correcting the default value of the SCAN system using sounding data from Taiwan to address these problems.</p> <p> </p>

Prediction of production indicators of fractured-vuggy reservoirs based on improved Graph Attention Network

Traditional forecasting methods rely on the temporal relationships within historical data, but they often require stationarity in the data, making them less effective for accurately predicting today's highly stochastic and volatile electricity loads. To address these limitations, this paper introduces an enhanced short-term electric load forecasting model: CNN-BiLSTM-Attention. Specifically, to overcome the spatial processing limitations of BiLSTM, a Convolutional Neural Network (CNN) module is incorporated, enhancing the model's ability to handle spatial aspects effectively. Additionally, to enhance the model's capability to capture important information features, an Attention mechanism is introduced, boosting its sensitivity to critical feature extraction. The proposed CNN-BiLSTM-Attention model integrates the strengths of different deep learning techniques, enabling it to comprehensively and accurately process complex datasets, particularly those requiring consideration of both local characteristics and long-term dependencies, as well as spatial considerations. The model achieved a remarkably high coefficient of determination <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(\mathrm{R}^{2})$</tex> of 0.99983, indicating excellent explanatory power and fit for electricity load data. The root mean square error (RMSE) was low at 19.2701, and the mean absolute error (MAE) was 12.8065, demonstrating minimal deviation between predicted and actual electricity load measurements, and thus high prediction accuracy. Compared to existing models, the proposed approach improved <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{R}^2$</tex> by 0.24% to 1.367%, reduced RMSE by 2.722 to 192.3089, and decreased MAE by 24.6896 to 185.3025. These improvements substantiate the model's significant advantages in enhancing prediction accuracy and minimizing errors. The model is well-suited to handle the uncertainties and non-linearities inherent in real-world electric load forecasting, meeting the pressing need for high precision, real-time, and stable forecasting capabilities, and providing a powerful tool for efficient electricity system operation.

Population growth in urban areas drives people vulnerability to natural disasters, especially in emergent economies like Brazil. The existing automatic pluviometric network is insufficient to provide enough information to monitor and prevent deaths due to landslides and floods in 821 municipalities defined as priority ones in the Brazilian National Plan of Risk Management and Response to Natural Disasters, implemented in 2012. Therefore, aiming to support early warning and monitoring systems and make available near-real time rain data, the National Early Warning and Monitoring Centre of Natural Disasters (Cemaden) has carried out an integrative institutional project to acquire and identify optimal places to install 1400 automatic pluviometers in risk areas of landslides and/or floods, already mapped in more than 350 Brazilian municipalities where existing network is scarce. Using geo-information techniques, the methodology included a deficit analysis to estimate the needed number of pluviometers per municipality. These estimates were used in a cooperative action between Cemaden and institutions at state and municipality levels, especially municipality governments and their civil defenses. A number of technical meetings and fieldwork training were done in 14 from the 27 Brazilian Federal Units in several municipalities to promote the project. Cemaden developed a dedicated web database where the institutions' representatives were able to supply detailed information of suitable sites to install pluviometers inside or nearby risk areas, based on their empirical knowledge of local characteristics. The results so far are very promising as 1, 500 locations have been filled since March 2013, and around 600 were chosen as optimal installation sites using spatial analysis. As the pluviometers installations take place, we expect a significant improvement on the efficiency and confidence of early warnings of landslides and floods as well as more commitment of local governments and communities to respond to warnings.

This letter proposes a novel, visually steered Reconfigurable Intelligent Surface (RIS) beam tracking method to enable rapid beam steering actions for high-mobility users. The approach utilizes advanced machine vision algorithms driven by artificial intelligence models, incorporating a specially designed 2-bit phase-quantized reflective metasurface and an embedded system control module. In this architecture, the RIS maintains line-of-sight visibility to both the base station and user, while the base station and user themselves remain in a non-line-of-sight configuration. Positioning the vision sensor at the RIS offers comprehensive environmental awareness, facilitating real-time user localization and motion prediction capabilities that are unattainable at either the transmitter or the user. The RIS dynamically updates codebooks to maintain stable links between transmitter and user without relying on feedback. Experiments confirm that the system achieves milliseconds-level tracking and reliable communication, offering a low-overhead solution for highly dynamic scenarios.

Ball sports have great variability in the game and the intelligent control of the rules of ball movement can effectively improve the training effect of athletes. However, the current research on artificial intelligence of spherical motion trajectory prediction points is basically blank. Based on this, this study is based on deep learning technology, and obtains the main experimental data through network data collection in the research and builds the table tennis spatial position image data set under various environments with accurate annotation based on the traditional deep learning. At the same time, the convolutional neural network is used as the location recognition algorithm, and a prediction algorithm for predicting the trajectory of table tennis is proposed based on the recurrent neural network. In addition, this paper designs comparative experiments to analyze the effectiveness of the algorithm model, and evaluates the real-time recognition, location and trajectory prediction capabilities, and conducts quantitative analysis. The research shows that the algorithm has certain practical effects and can provide theoretical reference for subsequent related research.

Border Gateway Protocol (BGP) serves as the primary routing protocol responsible for exchanging routing information between Internet Autonomous Systems (AS). Accurate BGP performance prediction is crucial for maintaining network stability and optimizing routing decisions. With the rapid evolution of the Internet, modern network architectures have become increasingly complex, exhibiting heterogeneous and highly dynamic characteristics. Networks encompass diverse devices, protocols, and inter-AS connections, making it challenging for traditional BGP performance prediction models to adapt to complex network environments. Traditional models often rely on predefined assumptions, such as static AS relationships, which prove inadequate when confronting constantly changing network topologies, fluctuating traffic patterns, and variable BGP routing decisions in realworld networks. Moreover, the heavy dependence of traditional models on specific data structures and network relationships further limits their generalization capabilities in broader applications.To address these challenges, this paper proposes a novel Deep Neural Network (DNN)-based model for predicting BGP convergence time using historical routing data from sources such as RouteViews and RIPE NCC. Our model employs comprehensive feature engineering, Principal Component Analysis (PCA) for dimensionality reduction, and Dropout regularization to effectively capture complex nonlinear relationships in BGP data. Experimental results demonstrate that the DNN model outperforms traditional methods, achieving an Area Under the Curve (AUC) value of 0.83, with accuracy trends and Receiver Operating Characteristic (ROC) curves validating its superior performance. Future research will incorporate temporal dynamics through Recurrent Neural Networks (RNN) to enhance real-time performance prediction capabilities in dynamic network environments.

ABSTRACTConsiderable effort has been devoted to developing real-time traffic network management systems for congestion mitigation in large metropolitan areas. These systems usually adopt high-resolution simulation models to provide real-time traffic network state estimation and short-term prediction capabilities. If the simulation results deviate from their corresponding real-world observations, online adjustment to the parameters of the simulation model is recommended to maintain its consistency. For that purpose, several consistency checking and online adjustment modules could be integrated with the simulation model and activated periodically to maintain the model consistency. This paper presents a multi-agent learning methodology for consistency checking and online calibration of real-time traffic network simulation models. The methodology allows multiple online adjustment modules to learn based on their historical performance. Consequently, an adjustment module is activated only if its activation is expected to reduce any detected model inconsistency. The methodology eliminates computation burden associated with the unnecessary activation of these adjustment modules, which might affect the system’s real-time execution requirement. The performance of the methodology is examined using real-world data. The results show that the methodology is promising as an efficient mechanism for maintaining model estimation consistency.

Flood forecasting is essential for protecting lives, infrastructure, and enabling timely disaster response. However, conventional centralized machine learning models struggle with data privacy, communication overhead, and limited access to distributed hydrological datasets. To overcome these challenges, this research proposes a Federated Deep Learning Framework for Real-Time Flood Prediction and Early Warning Systems, integrating advanced hydrological modeling with privacy-preserving federated learning techniques.Machine learning and deep learning methods have already demonstrated high potential for flood forecasting and detection [1], [7], [10], [19]. Time-series models such as LSTM provide strong capabilities for hydrological prediction [6], [11], while deep architectures like CNNs and ResNet enhance spatial feature extraction from remote sensing data [8], [18]. Federated learning enables decentralized model training without sharing raw data, ensuring privacy and security [3], [5], [12], [13], [20]. Further, optimization techniques like Adam and communication-efficient distributed learning approaches significantly improve model performance and scalability [15], [17], [23].The proposed framework combines rainfall data, river discharge, remote sensing imagery, and IoT-based flood monitoring inputs [21] to provide accurate and real-time flood prediction. Privacy-preserving federated aggregation prevents data leakage risks [16], while enabling collaborative learning across multiple geographically distributed nodes. Experimental evaluations show that federated deep learning improves accuracy, robustness, and scalability compared to centralized approaches, advancing toward a secure and intelligent flood forecasting ecosystem suitable for climate-sensitive regions.