Earth Science Informatics最新文献

Deep learning-based methods are widely used in hyperspectral image (HSI) classification and have achieved excellent classification performance. However, hyperspectral data from different categories exhibit strong nonlinear coupling, which results in low spatial distinguishability between samples from different categories. Under the condition of limited sample size, how to extract spectral-spatial features and reduce the coupling of hyperspectral data from different categories is the key to achieving high-precision classification. Some methods based on Convolutional Neural Networks (CNN) tend to focus on local information within hyperspectral cubes. Transformers have excellent performance in modeling global dependencies between sequences. To solve the above problems, this paper proposes a global local feature fusion network (GLF2Net) for hyperspectral classification. To effectively integrate global information, this method introduces frequency domain statistical methods into the field of hyperspectral image classification. Firstly, this paper utilizes Fast Fourier Transform (FFT) to obtain frequency domain information from HSI data. Then, an improved adaptive 13-dimensional frequency domain statistical feature is applied as a supplement to the information after Principal Component Analysis (PCA) dimensionality reduction. To fully capture local-global hyperspectral features from HSI data, a dual-branch structure with a Transformer encoder Convolution Mixer Branch (TCM) and a CNN Branch is designed. Through extensive experiments on real HSI datasets, it is proven that the classification performance of GLF2Net is superior to several classic HSI classification methods.

The Qanat (also known as kariz) is one of the significant water resources in many arid and semiarid regions. The present research aims to use machine learning techniques for Qanat discharge (QD) prediction and find a practical model that predicts QD well. Gene expression programming (GEP), artificial neural network (ANN), group method of data handling (GMDH), least-square support vector machine (LSSVM) and adaptive neuro-fuzzy inference system (ANFIS), are employed to predict one-, two-, and five-months time-step ahead QD in an unconfined aquifer. QD for one, two, and three lag-times (QD_t−1, QD_t−2, QD_t−3), QD for adjacent Qanat, the main meteorological components (T_t, ET_t, P_t) and GWL for one, two, and three lag-times are utilized as input dataset to accomplish accurate QD prediction. The GMDH model, according to its best results, had promising accuracy in predicting multi-step ahead monthly QD, followed by the LSSVM, ANFIS, ANN and GEP, respectively.

Seismic facies analysis which is aimed at identifying subsurface geological features from seismic data, has evolved due to the time-consuming and labor-intensive nature of its traditional approach. To address these challenges, numerical frameworks such as machine learning have been applied, yet attribute selection still comes with some challenges, particularly for inexperienced interpreters. Additionally, validating results in regions with limited well data poses significant challenges. This paper addresses these challenges through a comprehensive review of seismic facies workflows and a proposed workflow for a case study in the Gulf of Guinea. In this case study, seismic attribute selection is significantly based on the contribution (weights) of the individual attributes in a larger set of attributes. Also, we have introduced spectral decomposition for interpretation and initial validation of the workflow due to its independence on well data. Here, we applied an unsupervised vector quantizer to seismic attribute selection and facies analysis. Using a backward feature selection (BFS) approach for attribute selection based on computed weights assigned by our unsupervised vector quantizer (UVQ) network, we selected six seismic attributes for our facies analysis and tested five different attribute combinations of the attributes for facies analysis. This was followed by spectral decomposition colorblend of 5 Hz, 10 Hz, and 15 Hz frequencies. The facies generated using our seismic attributes varied with each combination due to the variations in the individual attributes. Correlating our seismic attributes and spectral decomposition to our facies, it was possible to identify lithological variations without solely relying on well data. Insights from this paper show the suitability of the automatic approach to seismic facies analysis in aiding the identification of new reserves which can bolster the economies of developing countries.

This study addresses the critical question of predicting the amplitude of S-waves during earthquakes in Aotearoa New Zealand (NZ), a highly earthquake-prone region, for implementing an Earthquake Early Warning System (EEWS). This research uses ground motion parameters from a comprehensive dataset comprising historical earthquakes in the Canterbury region of NZ. It explores the potential to estimate the damaging S-wave amplitude before it arrives, primarily focusing on the initial P-wave signals. The study establishes nine linear regression relationships between P-wave and S-wave amplitudes, employing three parameters: peak ground acceleration, peak ground velocity, and peak ground displacement. Each relationship’s performance is evaluated through correlation coefficient (R), coefficient of determination (R²), root mean square error (RMSE), and 5-fold Cross-validation RMSE, aiming to identify the most predictive empirical model for the Canterbury context. Results using a weighted scoring approach indicate that the relationship involving P-wave Peak Ground Velocity (Pv) within a 3-second window strongly correlates with S-wave Peak Ground Acceleration (PGA), highlighting its potential for EEWS. The selected empirical relationship is subsequently applied to establish a P-wave amplitude (Pv) threshold for the Canterbury region as a case study from which an EEWS could benefit. The study also suggests future research exploring complex machine learning models for predicting S-wave amplitude and expanding the analysis with more datasets from different regions of NZ.

Remote sensing images exhibit significant variations in target scale and complex backgrounds, as well as distinct differences within classes and high similarities between classes. These characteristics present particular challenges for remote sensing scene classification tasks. To address these issues, this paper proposes an efficient system architecture, the self-correction feature enhancement fusion network (SCECNet), designed to improve scene image processing capabilities. First, a feature pyramid network (FPN) based on ResNet50 is employed as the backbone for feature extraction, which helps alleviate feature loss for small targets. Second, a novel lightweight channel attention mechanism is designed to reduce the differences between features from different layers while suppressing irrelevant information. Next, a self-correction feature fusion module (SCFF) is constructed to further emphasise the main targets in complex environments through adaptive weighting. Finally, the classifier performs the final scene classification. Furthermore, a regional dataset, AHNR-18, is constructed to validate the generalisation capability of SCECNet and supplement existing datasets. Experiments on two benchmark datasets show that our method outperforms several existing methods.

This research aimed to investigate the effectiveness of Polyethylene-Terephthalate (PET) as a reinforcement material for sandy soils in enhancing the shear strength. To achieve this, different concentrations of PET were tested, and 118 sets of data were collected. Parameters such as relative density, normal stress in direct shear strength test, and types of PET elements (1 × 1, 1 × 5, and fiber) were also recorded. Subsequently, four decision tree-oriented machine learning (ML) methods—decision tree (DT), random forest (RF), AdaBoost, and XGBoost—were applied to construct models capable of forecasting enhancements in shear strength. The evaluation of these models' effectiveness was conducted using four established statistical metrics: R², RMSE, VAF, and A-10. The results showed that AdaBoost results in the highest prediction accuracy among other algorithms, representing the high modelling performance of the algorithm in dealing with complex nonlinear problems. The conducted sensitivity analysis also revealed that relative density is the most crucial parameter for all the algorithms in predicting the output, followed by PET percentage and normal stress. Furthermore, to make the developed model in this study more practical and easy to use, a Graphical User Interface (GUI) was created, enabling the engineers and researchers to perform the analysis straightforwardly.

This research presents a comprehensive drought analysis using climate data obtained from 219 homogeneously distributed meteorological stations in Türkiye between 1991 and 2022. In this context, Standard Precipitation Index (SPI), Standardized Precipitation Evapotranspiration Index (SPEI) and Reconnaissance Drought Index (RDI) drought indices were used and comparative analysis was made. Türkiye. The study demonstrates that below-normal precipitation over extended periods and increasing temperatures have contributed to the increased frequency of meteorological drought events. Türkiye's topographic conditions, particularly its location in the Mediterranean basin, significantly influence drought occurrences. It is noted that over the past 20 years, Türkiye has been trending towards drier conditions, with rising temperatures reinforcing this trend. The study observes that the moderate drought class range is the most frequently recurring in the SPI, SPEI, and RDI methods utilized. Regarding atmospheric conditions affecting the climate in Türkiye, it is observed that increased drought severity stands out prominently in years when the North Atlantic Oscillation is positive. During these years, increased drought severity is evident in the SPI, SPEI, and RDI indices, particularly in winter and autumn, while a wide area experiences drought effects in the summer months. Long-term analyses emphasize that drought periods occur less frequently but have more prolonged impacts, attributed to variations in precipitation patterns from year to year and the influence of rising temperatures due to global climate change. The potential future increase in drought in the Mediterranean basin due to global climate change and Türkiye's vulnerability to this situation could have adverse effects on water resources, food security, energy sources, and ecosystems.

The present study investigates the ultimate bearing capacity (UBC) of a footing subjected to an eccentric load situated above an unlined horseshoe-shaped tunnel in the rock mass, following the Generalized Hoek-Brown (GHB) failure criterion. A reduction factor (R_f) is introduced to investigate the impact of the tunnel on the UBC of the footing. R_f is determined using upper and lower bound analyses with adaptive finite-element limit analysis. The study examines the influence of several independent variables, including normalized load eccentricity (e/B), normalized vertical and horizontal distances (δ/B and H/B) of the footing from the tunnel, tunnel size (W/B), and other rock mass parameters. It was found that all these parameters significantly affect the behavior of tunnel-footing interaction depending on the range of varying parameters. The findings of the study indicate that the critical depth (when R_f is nearly 1) of the tunnel decreases with increasing load eccentricity. The critical depth is found to be δ/B ≥ 2 for e/B ≤ 0.2 and δ/B ≥ 1.5 for e/B ≥ 0.3, regardless of H/B ratios. Additionally, the GHB parameters of the rock mass significantly influence the interaction between the tunnel and the footing. Moreover, this study identifies some typical potential failure modes depending on the tunnel position. The typical potential failure modes of the footing include punching failure, cylindrical shear wedge failure, and Prandtl-type failure. This study also incorporates soft computing techniques and formulates empirical equations to predict R_f using artificial neural networks (ANNs) and multiple linear regression (MLR).

Flash floods rank among the most catastrophic natural disasters worldwide, inflicting severe socio-economic, environmental, and human impacts. Consequently, accurately identifying areas at potential risk is of paramount importance. This study investigates the efficacy of Deep 1D-Convolutional Neural Networks (Deep 1D-CNN) in spatially predicting flash floods, with a specific focus on the frequent tropical cyclone-induced flash floods in Thanh Hoa province, North Central Vietnam. The Deep 1D-CNN was structured with four convolutional layers, two pooling layers, one flattened layer, and two fully connected layers, employing the ADAM algorithm for optimization and Mean Squared Error (MSE) for loss calculation. A geodatabase containing 2540 flash flood locations and 12 influencing factors was compiled using multi-source geospatial data. The database was used to train and check the model. The results indicate that the Deep 1D-CNN model achieved high predictive accuracy (90.2%), along with a Kappa value of 0.804 and an AUC (Area Under the Curve) of 0.969, surpassing the benchmark models such as SVM (Support Vector Machine) and LR (Logistic Regression). The study concludes that the Deep 1D-CNN model is a highly effective tool for modeling flash floods.

Acquiring global ocean digital elevation model (DEM) is a forefront branch of marine geology and hydrographic survey that plays a crucial role in the study of the Earth’s system and seafloor’s structure. Due to limitations in technological capabilities and surveying costs, large-scale sampling of ocean depths is very coarse, making it challenging to directly create complete ocean DEM. Many traditional interpolation and deep learning methods have been applied to reconstruct ocean DEM images. However, the continuity and heterogeneity of ocean terrain data are too complex to be approximated effectively by traditional interpolation models. Meanwhile, due to the scarcity of available data, training an sufficient network directly with deep learning methods is difficult. In this work, we propose a conditional generative adversarial network (CGAN) based on transfer learning, which applies knowledge learned from land terrain to ocean terrain. We pre-train the model using land DEM data and fine-tune it using ocean DEM data. Specifically, we utilize randomly sampled ocean terrain data as network input, employ CGAN with U-Net architecture and residual blocks to capture terrain features of images through adversarial training, resulting in reconstructed bathymetric terrain images. The training process is constrained by the combined loss composed of adversarial loss, reconstruction loss, and perceptual loss. Experimental results demonstrate that our approach reduces the required amount of training data, and achieves better reconstruction accuracy compared to traditional methods.