To address the limitations of traditional RAIM algorithms, including abnormal fault detection and insensitivity to small faults, we introduce the concept of variance inflation from robust estimation into receiver autonomous integrity monitoring(RAIM). A RAIM algorithm combing variance inflation model is proposed, with an improved fault handling process. Integrity determination is performed using the user's protection level based on positioning uncertainty, and multi-constellation simulations and real-world tests in complex environments are conducted. The simulation results show that the proposed method can accurately detect a significant fault of 30 meters and effectively mitigate the impact of a small 5-meter fault on positioning results, successfully restoring the positioning error level to the state before the fault is introduced. The real-world test results show that the proposed method achieves horizontal and vertical availability rates of 95.38% and 98.74%, respectively, which are 11.63% and 5.06% higher than traditional RAIM algorithms. In terms of positioning accuracy, compared to traditional RAIM algorithms, the proposed method improves precision in E, N, U directions by 23.03%, 11.68%, and 26.79%, respectively.

Seamounts are previously approximated as an axisymmetric Gaussian shape, and the vertical gravity gradient(VGG) anomalies combined with ship-track bathymetry data are used to evaluate seamount morphology. However, there are significant challenges in describing the complex morphology of seamounts. Therefore, we classify seamounts into two distinct types: conical and elliptical conical shapes, which can be determined by three and five parameters, respectively. We perform a grid search on seamount parameters, obtaining VGG anomalies corresponding to each parameter through forward calculation. The forward VGG anomalies are compared with the observed VGG anomalies, and the minimal root mean square(RMS) error between them is used as a search criterion for estimating. We validate the effectiveness of grid search algorithm through numerical simulations, and apply it to practical examples. The estimation results are compared with multi-beam bathymetry data, the RMS errors ranging from 150 m to 640 m, and the fitting of VGG anomalies is within 8 Eötvös.

We propose a bathymetric prediction method based on a dual-channel BP neural network, integrate gravity anomaly data(including long-wave, short-wave, and residual gravity anomalies) and shipborne bathymetry data to predict seafloor topography in the gulf of Guinea. The prediction results are compared with those obtained from gravity geology methods and the GEBCO_2023 and Topo_25.1 models. The results show that dual-channel BP neural network-based inversion achieves superior accuracy compared to other models. This study can provide new insights into applying deep learning in marine geodesy and enhances the accuracy of seafloor topography inversion.

This article utilizes observational data from seven geoelectric field stations in the Shandong region starting in 2021. It determines the dominant azimuth angle α of the geoelectric field using a geoelectric rock mass fracture model. The jumping range Δα did not meet the criteria for judgment and did not meet the abnormal indicators. According to the prediction rules of this method, it is analyzed that the probability of a moderate to strong earthquake occurring in Shandong region in the near future is not high. Based on the analysis of the development of rock fractures in the site reflected by the advantageous azimuth results, the results show that the rock fractures in Heze station located on the east side of the Liaokao fault zone and Zoucheng station located between the Cangshan-Nishan fault zone and Fushan fault zone are in the generation stage, which is a common background for the development of small and medium-sized earthquakes area. The rock fractures of Anqiu station and Lingyang station located in the northern section of the Yishu fault zone and Rushan station on the east side of the Muping-Jimo fault zone are in the stage of development or growth, which is in line with the background of nurturing moderate to strong earthquakes and deserves special attention.

Aiming at the problems of wireless node seismometer can not be accurately located and may be lost in the field under complex exploration environment, the study of UHF RFID high-precision ranging and positioning is of great significance. Firstly, we use the received signal strength indicator(RSSI) approximation calculation formula to sieve out the sampling values with large errors. Then, two optimization objective functions of the third-generation non-dominated sorting genetic algorithm(NSGA-Ⅲ) are designed, whose independent variables are one-dimensional convolutional neural network(1D-CNN) hyper-parameters, such as learning rate decreasing factor, initial learning rate, and batch size, and whose dependent variables are determination coefficient(R²) and mean bias error(MBE) of network prediction and theoretical values. Finally, the best hyperparameter values are used to form a new model of NSGAⅢ-1D-CNN, which is designed to improve the stability and accuracy of RFID ranging model. The prediction experiments show that the new model has a small prediction error of node seismometer RFID ranging under 100 rounds of cyclic experiments, and its performance is excellent in several indexes, such as R², root mean square error(RMSE), mean absolute error(MAE), MBE, with the mean values of 0.977 9, 0.058 6 m, 0.047 2 m, -0.001 3 m, respectively, which has higher ranging and positioning accuracy compared with other models. The new model has certain application value in field physical exploration.

This study aims to examine the monitoring and early warning abilities of the Yunnan seismic station network using the 72 h continuous waveform recordings recorded at 2 060 stations. The evaluation index used was the maximum probability peak displacement(PGD) of background noise at each station. The findings revealed that the average monitoring capacity of the Yunnan seismic network raised to M_L1.0, while the smallest earthquake that could be monitored was M_L0.1, and the monitoring capacity was comparatively weak in the central and northwestern Yunnan area. Moreover, the average values of the minimum magnitude of earthquake early warning, the warning time after the triggering of the first station, and the warning time after the earthquake were M_L2.2, 4.0 s, and 6.1 s, respectively. Furthermore, the planning layout of the Yunnan seismic station network fulfilled the criteria for the initial establishment of earthquake early warning.

This study focuses on the classification problem of mining seismic and blasting signals in a mine in Gansu province. Common unsupervised classification methods such as support vector machine, K-means clustering, and principal component analysis were analyzed. These methods face significant challenges in terms of insufficient feature selection, low classification accuracy, and imbalanced samples. To attempt to address these issues, this paper employs a lightweight collaborative learning method(LCL-SSS) that combines MiniRocket feature extraction with improved K-means clustering, aiming to enhance the accuracy and efficiency of classification. Through time-frequency analysis, it was found that the initial P-wave frequency of blasting events is mainly concentrated between 80 to 200 Hz, while the P-wave frequency of mining seismic events is distributed between 20 to 120 Hz. After 30 training experiments, the classification accuracy of the LCL-SSS method reached 95.07%, effectively capturing key features in mining seismic waveforms.

Based on the PWV time series data inverted from GPS during 2010-2018 and the PWV data calculated from the ERA5 reanalysis data for the same period, a combination of the generative adversarial networks(GAN) model and the Transformer neural network model in deep learning was employed to achieve short-term forecasting of PWV in Greenland using GPS-PWV data. The prediction results were evaluated using the ERA5 data from 2019. The results show that the model performed well in most areas, with a root mean square error(RMSE) better than 4.5 mm and a correlation coefficient greater than 0.7. The correlation coefficients in spring, autumn, and winter were all above 0.5, while in summer, the correlation coefficients on some dates were slightly lower due to severe weather changes. This method is capable of predicting the spatial distribution and temporal variations of PWV in Greenland.

With the widespread use of GNSS on civil aviation and UAV platforms, scenarios with large elevation differences are increasingly emerging, which puts higher demands on the accuracy of vertical corrections for tropospheric delays. As the most advanced ZTD model currently available, GPT3 has limitations in its vertical corrections for temperature and pressure, making it only suitable for near-surface applications. To address this issue, this paper proposes an improved method of GPT3 by integrating a global temperature lapse rate grid model and a multi-variable atmospheric pressure-altitude formula, constructing the IGPT3 model, and evaluating its adaptability. The results show that compared with UNB3 and GPT3, IGPT3 has improved accuracy by 25.0%, 18.2%, 16.1%, and 72.2%, 50.0%, 69.9% in predicting temperature lapse rate, temperature profile, and pressure profile, respectively. Moreover, the precision improvements in predicting ZHD and ZTD profiles using IGPT3 reached 29.3 mm and 29.0 mm, respectively. Furthermore, the precision of GNSS water vapor inversion at most IGS stations was improved with the use of IGPT3, with the maximum improvement reaching 3.07 mm.

To assess the accuracy of Earth rotation parameters(ERP) derived from the current VLBI global observing system(VGOS) observational data, this paper processes data from VGOS observations made between 2019 and 2023. Simultaneously, traditional S-band and X-band VLBI observational data, which were obtained during the same periods as the VGOS data, were also included in the processing. The EOP 14 C04 series provided by the IERS were used as reference values for accuracy assessment. The results show that, during the selected experimental data periods, the average weighted root mean square(WRMS) of the post-fit residuals of time delay from VGOS observational data is 16.77 ps, which is significantly better than the 30.97 ps from the traditional S/X VLBI measurements during the same periods. In terms of ERP determination, there are still differences between the results from global network station observational data of VGOS and traditional S/X VLBI. Specifically, the ERP polar motion component accuracy derived from VGOS is inferior to that of traditional S/X VLBI, while the UT1-UTC accuracy is comparable to that of traditional S/X VLBI measurements. Further analysis using experimental stations centered on co-located stations indicates that VGOS measurements can provide ERP determination results with higher accuracy.