Cooperative spectrum sensing (CSS) in cognitive wireless sensor networks (CWSNs) is vulnerable to Byzantine attacks, resulting in reduced detection reliability. Although existing defense methods based on support vector machine (SVM) can enhance robustness, their parameter optimization relies on iterative cross-validation (CV), which incurs significant computational overhead and limits their application in resource-constrained environments. To address the efficiency bottleneck of SVM hyperparameter optimization in Byzantine-robust CSS, this letter proposes a sparse tent chaotic particle swarm optimization SVM (STCPSO-SVM) method. The novelty lies in integrating tent-chaotic particle initialization with a stagnation-aware sparse fitness update strategy, so that CV is invoked only for selected particles or during stagnation recovery rather than for the full swarm at every iteration. Simulation results demonstrate that, under composite attacks, compared with the optimal PSO-SVM, this method maintains a high detection accuracy whilst reducing computation time by approximately 24.9% and 53% for 220 and 500 training samples, respectively.

Cooperative spectrum sensing (CSS) in cognitive radio networks is vulnerable to Byzantine attack launched by malicious sensor nodes (MSNs), which threatens spectrum utilization efficiency. Therefore, this paper presents an enhanced sand cat swarm optimization (SCSO) algorithm for adaptively optimizing both the kernel parameter and penalty factor of support vector machine (SVM). The developed SCSO-SVM approach achieves fast and accurate MSN detection under multiple Byzantine attack scenarios. Simulation results show the proposed algorithm reduces the computational time by an order of magnitude against existing metaheuristic-based SVMs and sustains high accuracy with limited training samples.

Accurate rainfall nowcasting plays a vital role in flood control, agricultural planning, and climate research. However, the inherently nonlinear, highly variable, and spatiotemporally complex nature of precipitation poses considerable challenges for achieving reliable short-term forecasts. Conventional deep learning approaches, such as Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and Gated Recurrent Units (GRUs), often fail to capture the intricate temporal and spatial dependencies in precipitation data, particularly in regions with complex topography. To address these limitations, this study proposes AE-ConvGRU, an enhanced ConvGRU-based architecture that integrates a multi-head attention mechanism with feature enhancement strategies for early fusion of multimodal meteorological data, specifically Precipitable Water Vapor (PWV) and rainfall observations. The model adaptively assigns dynamic weights to highlight key spatial features and improve the representational capacity of fused data. Using ERA5 reanalysis datasets from 2019 to 2024 over Zhejiang Province, China, the AE-ConvGRU model was trained on multi-year data and evaluated on independent 2024 data. Six standard metrics, including the Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination R2, were used for performance assessment. Experimental results demonstrate that AE-ConvGRU achieved a one-hour rainfall prediction RMSE of 0.339 mm, representing a 27.10

Flood monitoring plays a critical role in supporting rescue operations and post-flood recovery efforts. The Cyclone Global Navigation Satellite System (CYGNSS) provides high-frequency, all-weather remote sensing observations that can be valuable for flood monitoring. However, due to the inherent spatial sparsity of CYGNSS data and the irregular distribution of its sampling, achieving both high temporal and high spatial resolution simultaneously remains challenging. Additionally, existing methods lack a reliable mechanism for assessing the accuracy of flood monitoring results. To address these limitations, an improved method based on the Neural-enhanced Spatial Interpolation model coupled with an Uncertainty Analysis Strategy (NIUA) is proposed. The core idea of this method is to enhance the spatial coverage of CYGNSS data through the Neural-enhanced spatial interpolation model, while simultaneously using the uncertainty analysis strategy to improve the accuracy and reliability of flood monitoring results. To evaluate the performance of the proposed method, data sets collected from July 18 to 26, 2021, in Zhengzhou, China are used. Experimental results demonstrate that, compared to conventional spatial interpolation method, the NIUA method leads to a significant improvement in interpolation accuracy. Specifically, the root mean square error (RMSE) decreases from 1.03 to 0.55, and the improvement rate is approximately 46.60%. In terms of the mean absolute error (MAE), it reduces from 0.47 to 0.27, and improvement rate is approximately 42.55%. Furthermore, the proposed method is also compared with other flood monitoring data sources, such as the Global Precipitation Measurement (GPM) and Soil Moisture Active Passive (SMAP) missions. Results show that the proposed method demonstrates a strong consistency with flood analysis derived from GPM and SMAP data. These findings validate the feasibility and superiority of the proposed method, highlighting its potential for improving the accuracy and reliability of flood monitoring applications.

The lack of robustness in multi-source integrated navigation systems under complex environments has become a widely discussed issue in recent years, especially in schemes involving magnetic fingerprinting as a key data source. To address this challenge, this paper proposes a robust optimization framework for integrated pedestrian dead reckoning (PDR), magnetic fingerprint vectors, and GNSS navigation. The framework introduces boundary constraints based on the inherent noise characteristics of sensors, combined with the hyperbolic tangent activation function, to ensure linear factor responses within constrained regions and avoid local extrema during iterative optimization. A robust loss function is applied to magnetometer and GNSS factors, effectively mitigating the impact of outliers. Additionally, a sliding window strategy with momentum-based optimization is employed to enhance performance in dynamic and complex scenarios. Simulation and field experiments demonstrate that the proposed framework achieves both high accuracy and strong robustness, with reductions in positioning errors of up to 46% compared to conventional factor graph approaches, meeting real-time positioning requirements even under satellite occlusion and electromagnetic interference.

ZTD (Zenith Tropospheric Delay) prediction plays an important role in GPS (Global Positioning System) positioning and weather monitoring. However, accurate ZTD prediction remains a major challenge due to the differences in the characteristics of its components, and its application is also limited in regions with sparse GPS stations. ZTD mainly consists of ZHD (Zenith Hydrostatic Delay) and ZWD (Zenith Wet Delay), whose fluctuation characteristics differ significantly from each other. Therefore, a parameter separation strategy is proposed to predict these two components separately by combining a period component extraction method with the Informer model to improve prediction accuracy and efficiency. Moreover, to obtain ZTD values at any location in regions with sparsely distributed GPS stations, a Graph Convolutional Network (GCN) method is adopted in this study. The performance of the proposed method is evaluated using datasets collected from 2013 to 2023 at eight GPS stations in Hong Kong, China. Experimental results show that: (1) after applying the parameter separation strategy, the MAE (Mean Absolute Error) and RMSE (Root Mean Square Error) of the predicted ZTD decreased from 8.77 mm to 7.61 mm and from 12.69 mm to 6.68 mm compared to the traditional method. The improvement rate is approximately 13.23% for MAE and 47.36% for RMSE, respectively. (2) The ZTD prediction accuracy at any point using the GCN method achieves about 8.80 mm for MAE and 12.88 mm for RMSE, which is comparable to that of the traditional method based directly on GPS observations. (3) The prediction accuracy of ZTD is strongly affected by seasonal variations. The MAE and RMSE are larger in the rainy season than in the non-rainy season. Specifically, the highest prediction accuracy occurs in winter, moderate accuracy in spring, and the lowest accuracy in summer and autumn. In conclusion, the proposed method effectively improves ZTD prediction accuracy and can be applied in regions with sparsely distributed GPS stations. (c) 2026 COSPAR. Published by Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

Cognitive radio (CR) technology aims to address spectrum scarcity by allowing sensor nodes (SNs) to detect and opportunistically access channels that are not utilized by primary users (PUs), thereby improving spectrum utilization. By applying cooperative spectrum sensing (CSS), CR improves spectrum detection accuracy, but also introduces security risks such as spectrum sensing data falsification (SSDF) attack. This letter proposes a defense strategy based on the improved isolation forest (IIF) for detecting malicious sensor nodes (MSNs) in cognitive wireless sensor networks (CWSNs). We analyze the distinct behavior patterns of the two types of SNs, and detect MSNs by isolating data points. Furthermore, we present an enhanced version of the traditional isolation forest to address its limitations. Finally, comparative experiments confirm the superiority of the proposed approach over six other anomaly detection algorithms, demonstrating the effectiveness of the improvements.

To address the decline in positioning accuracy caused by long-term global navigation satellite system (GNSS) observation outages under tightly coupling (TC) conditions in urban canyons, this article proposes a pseudorange compensation mechanism based on a bidirectional long short-term memory network with multiscale convolutional attention (BiLSTM-MSCA). Under frequent occlusion of satellite signals, the proposed network is used to learn the pseudorange incremental relationship between inertial navigation system (INS) information and GNSS signals, and then compensate for the GNSS pseudorange observations. The proposed BiLSTM-MSCA utilizes the bidirectional information of input INS and GNSS signals in the time domain and enhances the extraction of key information to improve the prediction accuracy of the network. Experiments based on the measured data of urban canyons show that the horizontal positioning accuracy of the proposed method is improved by 20% compared with the existing neural network-assisted method under the condition of 100-s GNSS observation loss.

Rainfall forecasting based on the GNSS (Global Navigation Satellite System) technique has the advantage of high temporal resolution and is one of the key technologies for rainfall monitoring. However, existing methods based on the GNSS technique rely heavily on historical data for establishing the forecasting model, which leads to the invalidation of these methods in areas where GNSS stations are lacking or sparse. To address these issues, a new method based on the multi-source data fusion and multi-dimension interpolation strategy is proposed in this study. The key idea of the proposed method can be divided into two parts. The first part: for areas lacking GNSS stations, the ERA5-PWV (The fifth generation of European Centre for Medium-Range Weather Forecasts-Precipitable Water Vapor) data sets are used to establish the forecasting model based on the LSTM (Long Short-Term Memory) model, then, the GNSS-PWV is directly used to forecast rainfall based on the trained forecasting model, which is denoted as the EPGP method. The second part: for areas with sparse GNSS stations, the multi-dimension interpolation strategy is adopted to obtain the GNSS-PWV in any area of the GNSS stations sparse network for rainfall forecasting, which is denoted as the MIGP method. Data sets collected from 2019 to 2022 in Hong Kong, China are used to evaluate the performance of the proposed method. Experimental results show that after adopting the proposed EPGP method, the TFR (True Forecasting Rates) and MFR (Miss forecast rate) of rainfall forecasting can reach approximately 90.12 % and 9.88 % for all rainfall types, which is comparable to the results based on the GNSS trained model. In terms of the MIGP method, the difference between the GNSS-PWV obtained from the multi-dimension interpolation method and obtained directly from the GNSS station is only 4.91 mm. Moreover, the TFR and FFR of the proposed MIGP method is approximately 89.83 % and 20.94 %, respectively. These results are equivalent to the method based on original GNSS data sets, which the TFR and FFR are 90.32 % and 20.54 %. Thus, from the above results, it can be concluded that the proposed EPGP and MIGP methods can effectively resolve the problems that GNSS rainfall forecasting technology cannot be used in the area of GNSS stations lacking or sparse. (c) 2025 COSPAR. Published by Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

Accurate rainfall forecasting plays a crucial role in weather monitoring. Currently, the application of global navigation satellite system-derived precipitable water vapor (GNSS-PWV) has mainly focused on forecasting rainfall event occurrence, while neglecting the forecasting of rainfall amount. In this study, a new method based on machine learning fusion strategy and the constraint of rainfall events is proposed. The machine learning fusion strategy is used to improve the accuracy of rainfall amount forecasting by considering the difference in rainfall types and machine learning algorithms, while the rainfall event constraint strategy is used to reduce the rainfall amount forecasting error during periods without rainfall event. In the new method, the long short-term memory (LSTM) algorithm is adopted to forecast rainfall events by considering the temporal correlation of rainfall events. Then, the support vector regression (SVR), group method of data handling (GMDH), and harmony search (HS) algorithms are combined to forecast rainfall amount with the constraint of rainfall event forecast results. The SVR performs better at forecasting little and medium rainfall, while GMDH is better at forecasting heavy rainfall. HS is mainly used to optimize the parameters of the forecasting model. The datasets collected from 2019 to 2023 at Hong Kong Global Navigation Satellite System (GNSS) stations are used to evaluate the performance of the proposed method. The experimental results show that after adopting the proposed rainfall event forecasting method, the total true positive rate (TPR) is approximately 90.45%, and the prediction errors for without rainfall events are only about 4.09%. Moreover, after adopting the proposed rainfall amount forecasting method, the average root mean square error (RMSE) and mean absolute error (MAE) are approximately 0.42 and 0.13 mm, respectively. The average improvement rate of MAE is approximately 24.39% for the internal experiment and 25.27% for the external experiment compared to the single SVR model. In conclusion, the proposed method can effectively improve the accuracy of rainfall amount forecast, which can further provide support for meteorological monitoring.

Accurate wind speed forecasting is crucial for the efficient integration of wind power into electrical grids, and integrating multi-scale wind speed features represents an effective approach to enhance prediction accuracy. However, existing models generally suffer from high computational costs and low efficiency when fusing multi-scale wind speed features. To address this challenge, this study proposes LASH-Net, a model capable of achieving excellent forecasting accuracy with low computational load. The model consists of a Long-term Historical Feature Extractor (LHFE) and a Short-term Historical Feature Extractor (SHFE), which extract coarse-scale and fine-scale features of wind speed sequences respectively. After integrating multi-scale information through feature concatenation, the final forecasting is generated. Experimental results demonstrate that LASH-Net achieves superior forecasting accuracy in 97.4 % of the 38 comparative scenarios across three datasets. Furthermore, the model exhibits low parameter counts and linear complexity, significantly improving the computational efficiency of multi-scale feature fusion.

The strength of a GPS carrier-to-noise ratio (CNR) signal is closely influenced by the multipath effect. This effect becomes more pronounced during flood events, as the reflection coefficient of water is significantly higher than that of dry soil. Consequently, the CNR measurements of GPS signals are impacted by floods. Based on this theory, the fluctuation of the CNR during a flood can be used to accurately monitor the process of a flood from occurrence to recession. Considering that the strength of the CNR largely depends on the satellite and frequency, and the characteristics of the influence of a flood on the CNR are different for each frequency, a new method based on the GPS dual-frequency direct signal CNR and the strength constraint threshold strategy was developed to increase the accuracy of the flood dynamics monitoring process. By using 64 MGEX (Multi-GNSS Experiment) stations distributed globally, an accurate direct-signal CNR threshold model of GPS dual-frequency was established. The threshold model demonstrated that the average difference of the direct-signal CNR, which is larger than 45 dB-Hz, between adjacent days at GPS L1 and L2 frequencies is 0.0659 dB-Hz and 0.0661 dB-Hz, respectively. Moreover, GPS real datasets in Zhengzhou city, China, from DOY (day of year) 199 to DOY 203, 2021, were collected to assess the proposed method. Based on the fluctuation of the direct-signal CNR threshold, the experimental results show that the flood appeared at about 16:04 PM on DOY 200, 2021, reached a peak at approximately 5:05 AM on DOY 202, and totally subsided at about 8:54 AM on DOY 202. Thus, the experiment results reveal that the proposed method accurately monitors the entire process of a flood from occurrence to recession, which provides valuable insights into operational flood dynamics, warning, and monitoring based on the GPS technique.

Rainfall forecasting based on the Global Navigation Satellite System (GNSS) plays an important role in GNSS-meteorology applications. However, traditional methods that rely on GPS-ZTD (Zenith Total Delay) are limited by the influence of ZHD (Zenith Hydrostatic Delay), while those based on GPS-PWV (Precipitable Water Vapor) depend heavily on external meteorological parameters. Moreover, most existing studies have not addressed the forecasting of rainfall duration. In this study, the feasibility of rainfall forecasting using GPS-ZWD (zenith wet delay) is demonstrated through theoretical analysis and correlation assessment. A new method is proposed for forecasting light, medium, and heavy rainfall events, as well as rainfall duration, directly based on GPS-ZWD. Data sets collected from 2010 to 2020 in Hong Kong are used to evaluate the accuracy of GPS-ZWD and establish a forecasting threshold model, while data from 2021 to 2022 are employed to validate the proposed method. Experimental results show that (a) the correlation coefficients between GPS-ZWD and ERA5-ZWD, and between GPS-ZWD and NCEP-ZWD, are 0.99 and 0.93, respectively; (b) the true rates for light, medium, and heavy rainfall event forecasting are 83.10%, 93.25%, and 100% based on GPS-ZWD. The average true rate is 88.28%, which is comparable to the methods based on GPS-PWV and better than GPS-ZTD methods; (c) for rainfall duration forecasting, the average true rates for light, medium, and heavy rainfall are 74.55%, 85.35%, and 96.29%, respectively. These results indicate that the proposed methods for rainfall event and rainfall duration forecasting based on GPS-ZWD demonstrate strong performance and hold significant potential for advancing GNSS-meteorology applications.

In some applications of Precipitable Water Vapor (PWV) measured by Global Navigation Satellite System (GNSS), we have to consider PWV vertical adjustments (PWVVAs) due to the height differences between the target sites and the GNSS sites. Thus, developments of global empirical models for PWVVAs deserve our attention. In this study, we found that the decrease factor of water vapor (2) can be also used as the decrease factor of PWV when we use the Smith's vertical variations of water vapor. Based on this conclusion, PWVVAs can be carried out by using the empirical values of 2 and atmospheric total pressure (or zenith hydrostatic delay). Thus, we gave two empirical models for PWVVAs (i.e. PWVVA-I and PWVVA-II). On the other hand, another aim of this research is to develop an empirical model of PWVVA using neural network (i.e. PWVVANN). Measured PWV of one site, the heights of this site and the target site, and empirical values of 2 and zenith hydrostatic delay were considered as the input parameters of the PWVVANN model. The output is the PWV at the target height. Global statistical results verify that the PWVVANN model has a accuracy of 1.08 mm and its accuracy has increased by respective 30.7 %, 24.4 % and 23.6 % when compared with PWVVA-I, PWVVA-II and a state of the art model (i.e. the GPWV-H model). (c) 2024 COSPAR. Published by Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

To meet the spectrum access requirements of unmanned aerial vehicles (UAVs), cooperative spectrum sensing (CSS) is utilized, allowing UAVs equipped with onboard sensors to collaboratively detect and access underutilized channels licensed to the primary user (PU). However, in UAV-assisted cognitive wireless sensor networks, CSS is highly susceptible to Byzantine attacks, which significantly degrade sensing performance. Furthermore, the collaborative sensing process often incurs high energy consumption and communication overhead among UAVs. To tackle these issues, this paper proposes a sequential local sensing strategy based on an improved voting rule, integrated with a sensing delay constraint mechanism, to enhance both robustness and energy efficiency in cognitive unmanned aerial vehicle networks (CUAVNs) under Byzantine attacks. Simulation results demonstrate that the proposed method not only mitigates the increase in error probability more effectively but also reduces energy consumption considerably, confirming its superior capability in balancing detection accuracy and energy conservation.

In this study, the influence of the South-to-North Water Diversion Project (SNWDP), China on meteorological parameters, including precipitable water vapor (PWV), precipitation, and three types of soil moisture, are first analyzed from 1994 to 2023 in the middle route flowing area. Three statistical methods based on the Generalized additive model (GAM), Theil-Sen trend analysis (TSTA), and Mann-Kendall significance test (MKST) are adopted to investigate the variation of these parameters. GAM and TSTA are mainly used to evaluate the variation trend from the perspective of temporal and spatial, respectively. MKST is used to assess whether the variation is significant. Through the comparative analysis before and after the opening of the SNWDP, as well as the overall period of analysis, three important findings can be concluded: (1) Before the opening of SNWDP, all of the PWV, precipitation, and soil moisture had continuously declined in the past 20 years from 1994 to 2014. The average decrease of PWV and precipitation is approximately 0.04 mm and 18.19 mm, respectively. Moreover, regarding the three types of soil moisture, the average volumetric water content decreased by 0.17

For cooperative spectrum sensing in cognitive wireless sensor networks, this paper proposes a weighted sequential probability ratio test (WSPRT) algorithm based on local consistency verification to mitigate Byzantine attack from malicious sensors. The method first calculates a local weight for each sensor by evaluating the consistency between its own sensing results and those of its neighbors. This local weight is then combined with the traditional WSPRT weight to produce a joint weighting. The objective is to identify and reduce the influence of malicious sensors on the fusion center. The fusion center finally makes a global decision on the primary user’s state using the combined weighted likelihood ratio. Simulation results demonstrate that the proposed method outperforms existing schemes in identifying malicious sensors and improving CSS reliability, requiring only about 3 samples on average to achieve a reliable global decision.

Spectrum sensing, a core element of cognitive radio (CR) technology, enables real-time monitoring of wireless spectrum usage to identify idle frequency bands and improve the efficiency of spectrum utilization. However, the open nature of cognitive wireless sensor networks (CWSN) makes collaborative spectrum sensing susceptible to Byzantine attacks from malicious sensor nodes (MSNs), which can severely degrade sensing performance and operational reliability. In response to the inherent instability of signal conditions and the threat of adaptive malicious attacks, this letter proposes an adaptive defense model based on density-based spatial clustering of applications with noise (DBSCAN), designed to counteract Byzantine attacks in environments with unknown and time-varying parameters. By integrating an enhanced DBSCAN algorithm with a dual decision-making mechanism-combining Kullback-Leibler (KL) divergence and energy thresholding-the model effectively identifies MSNs and determines the status of the primary node (PN) without relying on prior information. Simulations confirm that the proposed model maintains robust performance against Byzantine attack across diverse and dynamic radio conditions.

Factor graph models are widely used in multi-sensor fusion positioning, but traditional graph models still suffer from poor robustness. In magnetic fingerprint navigation, sensor anomalies can degrade positioning accuracy. To address the weak anti-interference ability of the combination of Pedestrian Dead Reckoning (PDR) and magnetic fingerprint matching, the low integration of the PDR and magnetic fingerprint models, and the reliance of existing robustness improvement schemes on human experience and high-stability sensors, an integrated robustness correction scheme is proposed. This scheme uses system states to detect outliers and correct the multi-sensor observation noise matrix based on sensor correlations. A parameter mapping scheme is derived to improve the model's solution capability in magnetic sparse areas, and a momentum optimization algorithm based on curvature information is applied to optimize the model. Simulation and field experiments verify that the method demonstrates strong robustness and can achieve high-precision PDR and magnetic fingerprint fusion navigation using low-cost sensors.

In recent years, WiFi has garnered increasing attention due to its fine-grained Channel State Information (CSI). Theoretically, more information from the transmitter implies greater potential for accurate localization. To enhance indoor positioning accuracy, this paper proposes an EfficientNetV2-based calibration localization system that integrates Physical Layer (PHY) information from CSI and Medium Access Control Layer (MAC) information from Received Signal Strength Indicator (RSSI), referred to as the Effrf system. The system first eliminates outliers using the Robust Random Cut Forest (RRCF) algorithm and then constructs an extended grid fingerprint database through radial basis function interpolation. In the offline phase, three-dimensional feature map fingerprints based on CSI amplitude, CSI phase, and RSSI are constructed, referred to as CSR feature map fingerprints. The EfficientNetV2 model is then trained to establish the correspondence between the fingerprints and their corresponding positions. During the online phase, to reduce data preprocessing time and ensure real-time positioning, the RRCF streaming anomaly detection method is employed to eliminate outliers. The EfficientNetV2 model, accelerated by GPU, is then used for position estimation. To further enhance accuracy, a random forest (RF) model is utilized to correct high-error localization results generated by the EfficientNetV2 model. Under indoor Line-of-Sight (LOS) and Non-Line-of-Sight (NLOS) conditions, the proposed system achieves average localization errors of 0.5962 m and 1.0306 m, respectively.