Artificial Intelligence in Geosciences最新文献

Earthquakes are classified as one of the most devastating natural disasters that can have catastrophic effects on the environment, lives, and properties. There has been an increasing interest in the prediction of earthquakes and in gaining a comprehensive understanding of the mechanisms that underlie their generation, yet earthquakes are the least predictable natural disaster. Satellite data, global positioning system, interferometry synthetic aperture radar (InSAR), and seismometers such as microelectromechanical system, seismometers, ocean bottom seismometers, and distributed acoustic sensing systems have all been used to predict earthquakes with a high degree of success. Despite advances in seismic wave recording, storage, and analysis, earthquake time, location, and magnitude prediction remain difficult. On the other hand, new developments in artificial intelligence (AI) and the Internet of Things (IoT) have shown promising potential to deliver more insights and predictions. Thus, this article reviewed the use of AI-driven Models and IoT-based technologies for the prediction of earthquakes, the limitations of current approaches, and open research issues. The review discusses earthquake prediction setbacks due to insufficient data, inconsistencies, diversity of earthquake precursor signals, and the earth's geophysical composition. Finally, this study examines potential approaches or solutions that scientists can employ to address the challenges they face in earthquake prediction. The analysis is based on the successful application of AI and IoT in other fields.

In recent years, there has been a growing interest in using artificial intelligence (AI) for rainfall-runoff modelling, as it has shown promising adaptability in this context. The current study involved the use of six distinct AI models to simulate monthly rainfall-runoff modelling in the Bardha watershed, India. These models included the artificial neural network (ANN), k-nearest neighbour regression model (KNN), extreme gradient boosting (XGBoost) regression model, random forest regression model (RF), convolutional neural network (CNN), and CNN-RNN (convolutional recurrent neural network). The years 2003–2007 are classified as the calibration or training period, while the years 2008–2009 are classified as the validation or testing period for the span of time 2003 to 2009. The available rainfall, maximum and minimum temperatures, and discharge data were collected and utilized in the models. To compare the performance of the models, five criteria were employed: R², NSE, MAE, RMSE, and PBIAS. The CNN-RNN model simulates the rainfall-runoff model in the Bardha watershed best in both the training and testing periods (training: R² is 0.99, NSE is 0.99, MAE is 1.76, RMSE is 3.11, and PBIAS is −1.45; testing: R² is 0.97, NSE is 0.97, MAE is 2.05, RMSE is 3.60, and PBIAS is −3.94). These results demonstrate the superior performance of the CNN-RNN model in simulating monthly rainfall-runoff modelling when compared to the other models used in the study. The findings suggest that the CNN-RNN model could be a valuable tool for various applications related to sustainable water resource management, flood control, and environmental planning.

Recently, machine learning (ML) has been considered a powerful technological element of different society areas. To transform the computer into a decision maker, several sophisticated methods and algorithms are constantly created and analyzed. In geophysics, both supervised and unsupervised ML methods have dramatically contributed to the development of seismic and well-log data interpretation. In well-logging, ML algorithms are well-suited for lithologic reconstruction problems, once there is no analytical expressions for computing well-log data produced by a particular rock unit. Additionally, supervised ML methods are strongly dependent on a accurate-labeled training data-set, which is not a simple task to achieve, due to data absences or corruption. Once an adequate supervision is performed, the classification outputs tend to be more accurate than unsupervised methods. This work presents a supervised version of a Self-Organizing Map, named as SSOM, to solve a lithologic reconstruction problem from well-log data. Firstly, we go for a more controlled problem and simulate well-log data directly from an interpreted geologic cross-section. We then define two specific training data-sets composed by density (RHOB), sonic (DT), spontaneous potential (SP) and gamma-ray (GR) logs, all simulated through a Gaussian distribution function per lithology. Once the training data-set is created, we simulate a particular pseudo-well, referred to as classification well, for defining controlled tests. First one comprises a training data-set with no labeled log data of the simulated fault zone. In the second test, we intentionally improve the training data-set with the fault. To bespeak the obtained results for each test, we analyze confusion matrices, logplots, accuracy and precision. Apart from very thin layer misclassifications, the SSOM provides reasonable lithologic reconstructions, especially when the improved training data-set is considered for supervision. The set of numerical experiments shows that our SSOM is extremely well-suited for a supervised lithologic reconstruction, especially to recover lithotypes that are weakly-sampled in the training log-data. On the other hand, some misclassifications are also observed when the cortex could not group the slightly different lithologies.

Geophysicists interpreting seismic reflection data aim for the highest resolution possible as this facilitates the interpretation and discrimination of subtle geological features. Various deterministic methods based on Wiener filtering exist to increase the temporal frequency bandwidth and compress the seismic wavelet in a process called spectral shaping. Auto-encoder neural networks with convolutional layers have been applied to this problem, with encouraging results, but the problem of generalization to unseen data remains. Most published works have used supervised learning with training data constructed from field seismic data or synthetic seismic data generated based on measured well logs or based on seismic wavefield modelling. This leads to satisfactory results on datasets similar to the training data but requires re-training of the networks for unseen data with different characteristics. In this work seek to improve the generalization, not by experimenting with network architecture (we use a conventional U-net with some small modifications), but by adopting a different approach to creating the training data for the supervised learning process. Although the network is important, at this stage of development we see more improvement in prediction results by altering the design of the training data than by architectural changes. The approach we take is to create synthetic training data consisting of simple geometric shapes convolved with a seismic wavelet. We created a very diverse training dataset consisting of 9000 seismic images with between 5 and 300 seismic events resembling seismic reflections that have geophysically motived perturbations in terms of shape and character. The 2D U-net we have trained can boost robustly and recursively the dominant frequency by 50%. We demonstrate this on unseen field data with different bandwidths and signal-to-noise ratios. Additionally, this 2D U-net can handle non-stationary wavelets and overlapping events of different bandwidth without creating excessive ringing. It is also robust in the presence of noise. The significance of this result is that it simplifies the effort of bandwidth extension and demonstrates the usefulness of auto-encoder neural network for geophysical data processing.

This research aim to evaluate hydro-meteorological data from the Yamuna River Basin, Uttarakhand, India, utilizing Extreme Value Distribution of Frequency Analysis and the Markov Chain Approach. This method assesses persistence and allows for combinatorial probability estimations such as initial and transitional probabilities. The hydrologic data was generated (in-situ) and received from Uttarakhand Jal Vidut Nigam Limited (UJVNL), and meteorological data was acquired from NASA's archives MERRA-2 product. A total of sixteen years (2005–2020) of data was used to foresee daily Precipitation from 2020 to 2022. MERRA-2 products are utilized as observed and forecast values for daily Precipitation throughout the monsoon season, which runs from July to September. Markov Chain and Long Short-Term Memory (LSTM) findings for 2020, 2021, and 2022 were observed, and anticipated values for daily rainfall during the monsoon season between July and September. According to test findings, the artificial intelligence technique cannot anticipate future regional meteorological formations; the correlation coefficient R² is around 0.12. According to the randomly verified precipitation data findings, the Markov Chain model has a success rate of 79.17 percent. The results suggest that extended return periods should be a warning sign for drought and flood risk in the Himalayan region. This study gives a better knowledge of the water budget, climate change variability, and impact of global warming, ultimately leading to improved water resource management and better emergency planning to the establishment of the Early Warning System (EWS) for extreme occurrences such as cloudbursts, flash floods, landslides hazards in the complex Himalayan region.

We propose a novel machine learning approach to improve the formation evaluation from logs by integrating petrophysical information with neural networks using a loss function. The petrophysical information can either be specific logging response equations or abstract relationships between logging data and reservoir parameters. We compare our method's performances using two datasets and evaluate the influences of multi-task learning, model structure, transfer learning, and petrophysics informed machine learning (PIML). Our experiments demonstrate that PIML significantly enhances the performance of formation evaluation, and the structure of residual neural network is optimal for incorporating petrophysical constraints. Moreover, PIML is less sensitive to noise. These findings indicate that it is crucial to integrate data-driven machine learning with petrophysical mechanism for the application of artificial intelligence in oil and gas exploration.

The vertical profile of the ionosphere density plays a significant role in the development of low-latitude Equatorial Plasma Bubbles (EPBs), that in turn lead to ionospheric scintillation which can severely degrade precision and availability of critical users of the Global Navigation Satellite System (GNSS). Accurate estimation of ionospheric delays through vertical electron density profiles is vital for mitigating GNSS errors and enhancing location-based services. The objective of this study is to propose a neural network, trained with radio occultation data from the COSMIC-1 mission, that generates average ionospheric electron density profiles during dusk, focusing on the pre-reversal enhancement of the zonal electric field. Results show that the estimated profiles exhibit a clear seasonal pattern, and reproduce adequately the climatological behavior of the ionosphere, thus presenting strong appeal on ionospheric error attenuation.

In recent years, convolutional neural networks (CNNs) have demonstrated their effectiveness in predicting bulk parameters, such as effective diffusion, directly from pore-space geometries. CNNs offer significant computational advantages over traditional methods, making them particularly appealing. However, the current literature primarily focuses on fully saturated porous media, while the partially saturated case is also of high interest for various applications. Partially saturated conditions present more complex geometries for diffusive transport, making the prediction task more challenging. Traditional CNNs tend to lose robustness and accuracy with lower saturation rates. In this paper, we overcome this limitation by introducing a CNN, which conveniently fuses diffusion prediction and a well-established morphological model that describes phase distributions in partially saturated porous media. We demonstrate the ability of our CNN to perform accurate predictions of relative diffusion directly from full pore-space geometries. Finally, we compare our predictions with well-established relations such as the one by Millington–Quirk.