pure and applied geophysics最新文献

Mapping subsurface stratigraphic patterns is crucial for enhancing the efficiency of oil and gas exploration and production operations. This study offers an approach to predict well-log data using machine learning (ML) techniques, focusing on the neutron log. Nine different ML algorithms were used to predict values of the neutron log: Multi-linear Regression, Hist Gradient Boost Regressor, AdaBoost Regressor, Decision Tree Regressor, Random Forest Regressor, Gradient Boost Regressor, Artificial Neural Networks (ANN), Bagging Regressor, and Light Gradient Boosting Machine (LightGBM). Each model was trained on three sets of input features (Series A, B, and C). To evaluate these models, the following parameters were used: Mean absolute error (MAE), Mean squared error (MSE), Root mean squared error (RMSE), Maximum error, Mean absolute percentage error (MAPE), and Adjusted R². In Series A, models were trained and tested using a dataset with two training features: Formation density (DENS) and Compressional Slowness (DTC). Series B models use three features: DENS, DTC, and Medium resistivity (RESM). Finally, Series C included five features: DENS, DTC, RESM, Gamma ray (GR), and Photoelectric effect (PEF). The comparative analysis of the model outcomes from the three feature sets, Series C models give the highest accuracy, with adjusted R² values up to 0.90, and lower error metrics such as RMSE as low as 0.03 and MAPE of 0.11. Series B models also showed good performance, with adjusted R² scores up to 0.82 and error values slightly higher than Series C this indicates that it could be a reliable alternative when fewer input features are available for model development. Overall, this study demonstrates three different series for neutron log prediction based on the accuracy of results with various training parameters and the data quality. Among the models evaluated, the Random Forest Regressor (Model 5) gives the best overall prediction performance, especially when provided with a wide-ranging and relevant input feature set.

In October 2020, a low-magnitude earthquake struck near the village of Capena, approximately 30 kilometers north of Rome. This seismic event was particularly remarkable due to its shallow hypocenter and the complete absence of aftershocks. Additionally, both during and after the event, residents reported hearing loud rumbling noises on multiple occasions in the village and surrounding rural areas. The earthquake's epicenter is located in a sinkhole-prone region intersected by major infrastructure, including a high-speed railway, high-voltage power lines, and a provincial road. Given these conditions, public authorities and regional Civil Protection agency closely monitored the area, prompting a comprehensive investigation. The study adopted an interdisciplinary approach to assess the stability of sinkhole-prone zones and identify any geological changes potentially triggered by the earthquake. A key focus of the research was the "Lago Puzzo," a well-documented example of multiphase, still active, sinkhole evolution within the San Martino Valley, alongside other similar extinct features in the area. The investigation integrated three primary methodologies: spectral analysis of seismic data to characterize the October 2020 event; remote sensing (SAR) using InSAR analysis to map ground displacement; geophysical surveys (ERT) to reveal subsurface structures. Although the study found no direct correlation between the earthquake and the evolution of Lago Puzzo, the results provide valuable insights into the current deformation state of the sinkhole and offer projections for its potential future development. Ultimately, this research represents a significant step forward in refining hazard assessment methods for the region.

The Piton de la Fournaise volcano (PdlF) on the island of La Réunion is one of the most active and best monitored volcanoes in the world. Its frequent eruptions make it a natural laboratory for developing new methods and evaluating their performance over multiple eruption sequences. In this work, we present a Deep Learning (DL) model for volcanic earthquake detection and two models for classification based on DL and Machine Learning (ML) algorithms. The detection model is based on encoder–decoder layers that extract high-order features in the time domain that are hidden in the seismograms. The first classification model consists of a simple convolutional neural network that uses the short-time Fourier transform of the signals as input data. A second classifier is based on ML approach and uses hand-crafted features. We show that our detection model, trained on ~ 7 000 volcano-seismic events recorded at PdlF between 2014 and 2021, outperforms previous DL-based models in detecting volcano-seismic events, achieving an accuracy of 98.15% on the testing dataset. Seven classes of signals are considered for classification models: volcano-tectonic (VT) events, rockfall, long-period events, volcanic tremors, tectonic events, anthropogenic noise and environmental noise. Both tested classification models achieve an accuracy of 96.55% in the testing dataset. By applying these models to the continuous data recorded at PdlF in 2019, we are able to detect and classify 1.5 times more VT events than the catalog provided by the Observatory. The detection model takes 28 s to process 24 h seismograms and from a few to a maximum of 70 s for classification.

Besides Earth tides, the atmosphere causes the most significant contributions to time-variable gravity. State-of-the-art modelling approaches of atmospheric loading (Newtonian attraction and deformation) benefit from numerical weather models in order to account for global air mass variations. Atmospheric loading effects are often computed assuming that the oceans respond as an Inverse Barometer (IB) which is not completely valid for periods shorter than a few weeks. Therefore, a more precise modelling is only possible considering the ocean response to atmospheric pressure and winds based on simulations of an ocean dynamic model. Superconducting gravimeters (SGs) measure temporal gravity variations with high-resolution and exceptional stability and are capable to sense mass redistribution in the atmosphere, the oceans and in terrestrial water storage. In this sense, although SGs provide information for particular sites, they allow for a reliable validation of mass variations represented by models for atmosphere and oceans. In the present study, a comparison of atmospheric and non-tidal ocean loading corrections as calculated by the recently updated Atmospheric attraction computation service (Atmacs) and the EOST Loading Service is performed. For this comparison, a set of SG stations is selected, with focus on stations close to the oceans where non-tidal ocean loading effects are more significant. An emphasis is made on the reduction of gravity residuals by different corrections.

The summer monsoon is a major weather phenomenon and has a significant socio-economic influence on India. In this context, understanding the long-term trend and the dynamics of extreme rainfall events (EREs), especially over the arid and semiarid regions like northwest India, remains limited. In this study, we delve into not only understanding the long-term trend and how they evolve in a warming climate but also elucidate the complex driving mechanisms of these events over the arid and semiarid regions of northwest India. Using high-resolution long-term rainfall datasets, EREs are classified into three major categories based on their spatial extent: widespread extreme rainfall events (WEREs), localized extreme rainfall events (LEREs), and EREs (comprising both WEREs and LEREs). Trend analysis spanning over a century indicates a significant upward trend in the frequency of WEREs. Additionally, the contribution of EREs to seasonal rainfall shows a notable increase, especially over western Gujarat. The composite analysis reveals distinct patterns in WEREs and LEREs, showcasing the evolution of rainfall and anomalous low pressure over the region. Results suggest strong vertical velocity and middle-level tropospheric heating are the key drivers of these EREs. In addition, high values of potential vorticity and moist static energy in the mid-level are key features of these systems, with snow hydrometeor being the major contributor to rainfall. This study's findings underscore the growing concern about EREs affecting the arid and semiarid region, which is already vulnerable due to their limited vegetation and scarce water resources.

In the geotechnical field, sediment layer thickness determination is critically important, as it provides invaluable information for structural and infrastructure planning and design. The Bakauheni area, characterized by numerous calderas and ancient volcanic deposits from the Pliocene to Holocene epochs, presents a compelling case for studying sediment layer thickness. Using 64 Horizontal-to-Vertical Spectral Ratio (HVSR) measurement points, we investigated sediment thickness distribution and its correlation with ancient calderas in Bakauheni. A 1D shear wave velocity (Vs) model was derived through inversion using the Particle Swarm Optimization (PSO) algorithm, revealing an average Vs of ~ 600 m/s across the study area. This relatively high Vs value suggests dense, compacted sediments or weathered bedrock. The average HVSR curve yielded a natural frequency (f₀) of 15.12 Hz, corresponding to an estimated sediment thickness of 9.92 m (assuming Vs = 600 m/s). This aligns closely with the median thickness of 10.55 m calculated from all 64 measurement points. Observed sediment thicknesses ranged from 4.39 to 103.57 m, with a mean of 18.22 m, indicating a general thickness range of 10–18 m in Bakauheni. The thickest deposits (> 30 m) correlate with caldera locations and low-topography zones, implying substantial sediment accumulation over ancient calderas. While the HVSR method effectively estimates sediment thickness for caldera identification, local Vs variations due to sediment composition necessitate further research to fully characterize the subsurface properties.

Vertical Electrical Sounding (VES) data inversion is a complex non-linear problem requiring robust global optimization methods. While techniques like Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) are widely used, the Flower Pollination Algorithm (FPA) offers distinct advantages: (1) faster convergence due to its Lévy flight mechanism, (2) handling the overfitting problem, and (3) superior balance between exploration and exploitation. In this study, we enhance FPA with adaptive switch probability to further optimize its performance for VES inversion. The adaptive switch probability in FPA dynamically adjusts the balance between global exploration (early iterations) and local exploitation (late iterations), improving convergence speed and accuracy compared to fixed-probability FPA. This is achieved by gradually reducing the switch probability p from = 0.8 to = 0.2 over iterations, optimizing the search process for VES inversion. Tests were conducted using synthetic VES data with 3-layer (Q-type and H-type) and 4-layer. The adaptive FPA demonstrated superior performance to conventional FPA, reducing misfit errors by 30–50% across synthetic models (e.g., 0.0011% vs. 0.0044% for H-type) while maintaining 40% faster convergence rates. Improvement was quantified using three key criteria: (1) final misfit error, (2) iteration count to convergence, and (3) parameter uncertainty, with the adaptive version consistently outperforming in all metrics. Adaptive switch FPA was also applied to VES field data for groundwater exploration in Leihitu village, Central Maluku, Indonesia. The adaptive FPA's subsurface reconstructions were validated through direct comparison with IP2WIN commercial software, showing superior resolution in identifying layer boundaries (≤ 5% deviation in thickness estimates) and resistivity values (≤ 8% relative error). In Leihitu village, the freshwater aquifer (18–22 Ωm) at 22–33.8 m depth sandwiched between impermeable limestone layers (> 300 Ωm) and the aquifer is sandstone which has good porosity. In the future, the FPA adaptive switch can be tested to solve complex non-linear geophysical inversion problems such as Controlled Source Audio-frequency Magnetotellurics (CSAMT), Magnetotelluric (MT), Transient Electromagnetic (TEM), and others.

Drought is not only a problem that challenges scientists but also one of the most difficult natural disasters to combat for local governments and decision-makers. Like many parts of the world suffering from drought, the western Mediterranean region of Turkey is also affected by drought. In this study, innovative drought prediction models were created with different machine learning algorithms and deep learning methods to create a model that will help decision-makers regarding drought. 4 different monthly lagged model structures were established using SPI12 values calculated with precipitation data between 1967 and 2020 for the Acipayam, Bodrum and Fethiye regions located in the west of Turkey. While providing data, attention was paid to the distance between stations and data continuity. The models were analyzed with Long-Short Term Memory (LSTM), Artificial Neural Network (ANN) and Random Forest (RF) algorithms. In addition, Discrete Wavelet Transform (WT) was used to obtain better model results. The hyper-parameters of these algorithms were determined by taking into account the most commonly used parameters in the literature. The analysis results were evaluated by correlation coefficient (R), root mean square error (RMSE), Nash–Sutcliffe Efficiency (NSE), and Combined Accuracy (CA). As a result of the comparison of these methods, the best results were obtained in the M01 model of LSTM, LSTM-WM01, for Acipayam after WT (r: 0.9910, NSE: 0.9733, RMSE: 0.1637, and CA: 0.1030). While the best prediction for Bodrum was obtained in LSTM-WM02 after WT (r:0.9657, NSE:0.9325, RMSE:0.3101, and CA: 0.1929), for Fethiye it was obtained in LSTM-WM02 having performance metrics r:0.9539, NSE:0.8973, RMSE:0.3689, and CA: 0.2359. It is expected that the results obtained with this study will help researchers and decision-making authorities on drought.

New tidal solutions from laser tracking of eight geodetic satellites and from a constellation of radar altimeters are combined to determine the complex Love number (k_2) for four lunar tidal constituents in the diurnal and semidiurnal bands. The tidal solutions for each data type must account for inconsistent prior Love numbers; the altimetry community has historically used elastic Love numbers. Use of the complex Love numbers recommended by current international conventions results in a small (order 4%) discrepancy between altimeter and tracking solutions for the degree-2 prograde spherical harmonics; this points to an anelastic Earth model that is too dissipative, with a phase lag slightly too large. Our estimated phase lag for (k_2) varies slightly across the tidal bands, from (0.228^{circ } pm 0.024^{circ }) for (hbox {O}_1) to a smaller (0.178^{circ } pm 0.020^{circ }) for (hbox {M}_2), with corresponding tidal Q rising from 250 to 320. Results for (hbox {N}_2) and (hbox {Q}_1) are consistent, but with much larger uncertainties. There is some interdependence on the values adopted for other Love and loading numbers, which we account for. A possibly important systematic error arises from seawater density, needed to relate ocean tidal elevations to gravitational Stokes coefficients. A constant mean density of 1035 kg (hbox {m}^{-3}) is used, but allowance for spatial variations in ocean density may be necessary.

This study classifies synoptic-scale atmospheric patterns associated with hailstorms in the Democratic People’s Republic of Korea (DPRK) using observational hail data (2010–2018) and multivariate statistical methods. Principal Component Analysis and k-means clustering identified four distinct regimes: (1) Summer Monsoonal Ridge-Trough (60.6% of hail days), characterized by a weak 500-hPa trough over northern DPR Korea, warm advection at 850 hPa, and high convective instability (CAPE > 1460 J kg⁻¹, lapse rate 26 K); (2) Autumn Cold Frontal (7.7%), driven by Siberian cold-air incursions and moderate CAPE (~ 1160 J kg⁻¹); (3) Spring Closed Low (16.6%), marked by a cyclonic low over northeast China and orographically enhanced uplift in northern highlands; and (4) Spring–Autumn Shortwave Trough (15.1%), featuring mobile mid-level troughs and frontal convergence. Discriminant Analysis validated the classification (93.4% accuracy), with key predictors including 500-hPa geopotential height and 850-hPa temperature loadings. Topography critically modulated hail genesis, amplifying updrafts in mountainous regions despite weaker synoptic forcing. These findings provide actionable insights for forecasting, emphasizing CAPE thresholds (> 1500 J kg⁻¹) and ridge-trough interactions as precursors to severe hail. The study establishes a transferable framework for hail prediction in topographically complex regions, addressing a critical gap in East Asian climatological research.