Earth Science Informatics最新文献

This groundbreaking study pioneers the application of state-of-the-art machine learning algorithms to predict pivotal water parameters, specifically pH, water depth, and salinity. Rigorously evaluating four leading algorithms (Random Forest Regressor, MLP Regressor, Support Vector Machine, and XGB Regressor) leveraging a substantial dataset and employing comprehensive metrics, including R², MSE, MAE, and cross-validation scores.

Results unequivocally demonstrate the exceptional performance of MLP Regressor and XGB Regressor, consistently outclassing other models in predicting pH, with remarkable R² values and minimal errors. MLP Regressor excels as the preeminent model for water depth prediction, while XGB Regressor leads in accurately predicting salinity. The study underscores the paramount importance of cross-validation in meticulously assessing model robustness and generalization capabilities.

A distinctive feature of this research lies in its innovative approach, incorporating geographic localization data (longitude, latitude, and altitude) as exclusive inputs for all models. This strategic integration showcases the algorithms' unprecedented ability to predict water parameters based solely on geographical coordinates, underscoring the transformative potential of machine learning in revolutionizing water resource management.

The implications extend far beyond its immediate focus, encompassing critical areas such as geophysical exploration, environmental monitoring, water quality management, and ecological conservation. By providing invaluable insights into the application of machine learning algorithms for predicting key water parameters, this study positions itself at the forefront of scientific contributions, setting a new standard for excellence in sustainable water resource utilization.

Global warming and climate change impacts intensify hydrological cycle and consequently unprecedented drought and flood appear in different parts of the world. Meteorological drought assessments are widely evaluated by the concept of standardized precipitation index (SPI), which provides drought classification. Its application is based on the probabilistic standardization procedure, but in the literature, there is a confusion with the statistical standardization procedure. This paper provides distinctive differences between the two approaches and provides the application of a better method. As a novel approach, SPI classification is coupled with fuzzy clustering procedure, which provides drought evaluation procedure based on two variables jointly, precipitation and temperature, which is referred to as the standard precipitation-temperature index (SPTI). The final product is in the form of fuzzy c-means clustering in five clusters with exposition of annual drought membership degrees (MDs) for each cluster and resulting objective function. The application of the proposed fuzzy methodology is presented for the long-term annual precipitation and temperature records from New Jersey Statewide records.

Mineral resource modeling is always accompanied by challenges. It is pivotal to increase accuracy and reduce modeling errors in resource modeling. This research aims at improving the resource modeling results using auxiliary variables for estimation and simulation processes. For this purpose, the Darreh-Ziarat iron ore deposit in the west of Iran is selected as a case study. The susceptibility obtained from the 3D inversion result of the magnetometry data is used as a secondary variable in the resource modeling. First, the Fe grade was estimated by utilizing simple kriging (SK) and sequential Gaussian simulation (SGS) techniques. Then, using the auxiliary variable, the Fe grade was estimated by the cokriging (CK) and sequential Gaussian co-simulation (SGCS) methods. Considering various cut-off Fe grades, the average grade of Fe and its resource (tonnage) were calculated, and their results were compared. The mean of kriging variance saw a decline from 0.81 in the SK method to 0.67 in the CK method. This slight decrease in variance can create a profound impact on the resource classification results. The results showed that the use of an auxiliary variable in resource modeling of Darreh-Ziarat led to a reduction in estimation error, an improvement in the classification of mineral resources, and an increase in the number of high-grade Fe blocks. Finally, Fe grade values at different elevation levels were calculated using the four mentioned methods. The results revealed a strong resemblance in shallow and deep parts, while the middle part, which is the high-grade zone, showed more differences.

There is a need for a comprehensive comparative analysis of spatiotemporal variations in surface water quality, particularly in regions facing multiple pollution sources. While previous research has explored the use of individual water quality indices (WQIs), there is limited understanding of how different WQIs perform in assessing water quality dynamics in complex environmental settings. The objective of this study is to evaluate the effectiveness of three WQIs (Canadian Council of Ministers of the Environment (CCME), National Sanitation Foundation (NSF) and System for Evaluation of the Quality of rivers (SEQ-Eau) and a national water quality regulation in assessing water quality dynamics. The pilot study area is the Acısu Creek in Antalya City of Turkey, where agricultural practices and discharge of treated wastewater effluents impair the water quality. A year-long intensive monitoring study was conducted includig on-site measurements, analysis of numerous physicochemical and bacteriological parameters. The CCME and SEQ-Eau indices classified water quality as excellent/good at the upstream, gradually deteriorating to very poor downstream, showing a strong correlation. However, the NSF index displayed less accuracy in evaluating water quality for certain monitoring stations/sessions due to eclipsing and rigidity problems. The regulatory approach, which categorized water quality as either moderate or good for different sampling sessions/stations, was also found less accurate. The novelty of this study lies in its holistic approach to identify methodological considerations that influence the performance of WQIs. Incorporating statistical analysis, artificial intelligence or multi-criteria decision-making methods into WQIs is recommended for enhanced surface water quality assessment and management strategies.

Hyperspectral imaging has appeared as a vital tool in remote sensing science for its efficacy in effectively delineating regions of interest. However, the classification of hyperspectral images (HSI) encounters notable challenges, including the high dimensionality of highly correlated bands and the scarcity of training samples. Addressing these challenges is very essential by determining the most relevant bands, as well as the utilization of unlabelled training samples. In response to these issues, this study presents an unsupervised framework based on an enhanced Mayfly Optimization Algorithm (MOA) in order to select the most informative spectral bands. The enhanced MOA effectively identifies informative bands by leveraging the random solutions to explore the global search space, and enhance the solution diversity. On the other hand, leveraging the best experiences to boost the local search, efficiently attaining optimal solutions. This balanced exploration-exploitation strategy ensures the algorithm’s robustness and effectiveness in addressing the optimization problem. Ultimately, the proposed approach is demonstrated at the pixel-level hyperspectral image classification using two machine learning classifiers: Random Forest and Support Vector Machine. Thorough experimentation carried out on three benchmark hyperspectral datasets consistently confirms the effectiveness of the proposed approach.

For a long time, elucidating the tectonic setting of unknown rock samples has been a focal point for geologists. Traditional methodologies for this purpose have been scrutinized increasingly due to their inherent limitations. In response to these challenges, this paper applies modern machine learning techniques to analyze the geochemical data of igneous rocks and improve understanding of tectonic settings. By employing a variety of machine learning models, including Decision Trees, K-Nearest Neighbors, Support Vector Machines, Random Forests, Extreme Gradient Boosting, and Artificial Neural Networks, and training with 23 features comprising nine major elements (SiO₂, TiO₂, Al₂O₃, CaO, MgO, MnO, Na₂O, K₂O, and P₂O₅) along with 14 trace elements (La, Ce, Pr, Nd, Sm, Eu, Gd, Tb, Dy, Ho, Er, Tm, Yb, and Lu), the study successfully distinguished between seven different tectonic settings. Among these models, Random Forest, Extreme Gradient Boosting, and Artificial Neural Networks demonstrated superior classification accuracy and recall rates, with accuracies of 0.85, 0.87, and 0.86, respectively. This validates the effectiveness and potential of machine learning technologies in distinguishing the tectonic settings of igneous rocks through their geochemical elements. To enable geologists and researchers to more accurately understand and predict the origins of igneous rocks without the need to master machine learning knowledge, a user-friendly software, GeoTectAI, has been developed.

The randomicity and fluctuation of the wind speed will influence the precision of the forecast. This paper presents a new method of combined wind speed forecast based on the two-level decomposition and weighted average, which can improve the accuracy of wind speed forecasting. First, the improved complementary ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) decomposition method is used to get different sub-sequences, and then the fuzzy entropy is used to judge the degree of confusion of the sub-sequences. In this paper, the autoregressive integrated moving average (ARIMA) model is used to predict the minimum fuzzy entropy. And the other subsequences are decomposed by backpropagation neural network (BPNN), variational mode decomposition (VMD) and predicted by nonlinear auto-regressive (NAR) and BP neural network with suitable weighting ratio for weighted average and particle swarm optimization- long short-term memory (PSO-LSTM) neural network respectively, and ultimately all the predicted values are superimposed to get the final prediction. Experiments are conducted using three datasets and eight comparison models to verify the validity of this model. The prediction analysis was carried out using the actual measured data of a wind farm in Inner Mongolia, and the results indicated that (1) using fuzzy entropy can effectively improve the prediction precision; (2) the prediction accuracy of the combined prediction method of neural network based on two-level decomposition was greatly improved and the prediction results were more reliable; (3) Decompose one of the subsequences with VMD, predict it with NAR and BP neural network, and choose appropriate weight ratio for weighted average prediction will achieve better prediction results; (4) the root mean square error (RMSE) of the hybrid model on the three wind speed datasets were 0.28777, 0.22786 and 0.17128, which are lower than the comparison values of other models. So, it is workable to use this hybrid model in wind speed prediction.

Over the past few years, forest ecosystems’ ability to store carbon has been significantly impacted by Land use and Land cover (LULC), and climate change. Thus, it is crucial to understand how these change-causing factors impact carbon sequestration (CS). Due to a limited number of carbon storage monitoring methods and the shorter period of remote sensing data, it is difficult to continually analyze carbon storage in large areas. These issues can be solved by using AVHRR (Advanced Very High-Resolution Radiometer) and MODIS (Moderate Resolution Imaging Spector radiometer) remote sensing data. The main objective of this research is to measure the spatial and temporal patterns of carbon storage across the state of Karnataka’s vegetative and non-vegetated terrains, between 2003 and 2021. To assess the effects of potential land use and land cover scenarios, our work uses spatial maps to estimate the storage of carbon sequestration from various land use patterns. To assess the spatio-temporal effects of land use and land cover (LULC) change on the availability and value of carbon storage. This research focuses on the entire Karnataka state as a study region to compute carbon storage utilizing online platforms like GEE (Google Earth Engine) using GPP (Gross Primary Productivity), and NPP (Net Primary Productivity) is an important measure to evaluate vegetation productivity using Decision Tree (DT) machine learning techniques. Statistical models like Pearson’s correlation coefficient, standardized coefficients, and Root Mean Square Error (RMSE) methods are used for the model’s performance with different indices and carbon storage. The findings show the Uttara Kannada district contains between 250 gCm − ² and 300 gCm − ² of carbon storage, which is relatively significant as compared to the other parts of the districts in the state.

In recent years, geoscientists have been employing machine learning techniques to automate lithological identification by integrating well logging data. However, in geologically complex regions, few have taken into consideration the differences between boreholes and the uneven distribution of lithology. Additionally, there has been limited effort to differentiate boreholes in the same region based on stratigraphic sequences when addressing these issues. We propose a workflow for machine learning-based automated lithological identification. Utilizing the Structural Deep Clustering Network (SDCN) algorithm for deep clustering, we differentiate logging sampling points with geological strata as the clustering scale, assigning each sampling point to its corresponding stratum. In order to obtain stratum information for each borehole, we have devised a Borehole Cluster Result Processing Layer. By segmenting logging data windows, we extract stratum information for each borehole, using the distinctiveness of borehole stratum information as the basis for borehole classification. Subsequently, we assess the impact of lithological classification on logging data for each borehole category using four machine learning methods: extreme gradient boosting (XGBoost), adaptive boosting (AdaBoost), bidirectional long short-term memory (Bi-LSTM), and bidirectional gated recurrent unit (Bi-GRU). The experimental results indicate that, compared to the case where boreholes are not classified, the lithological classification performance for the majority of borehole categories has improved by 1% to 6%. However, there is also a category of boreholes where the classification performance is less than ideal due to the significant variability of diabase contained within the Paleogene strata in the electrical resistivity logging.

This paper proposes EMA-Net, a fully convolutional neural network, to improve the effectiveness of sea-land segmentation on Gaofen-2 images. The aim is to address the issue of low segmentation accuracy in sea-land boundary regions when using remote sensing images for sea-land segmentation. The MA-Net network structure is enhanced by splitting the EfficientNet-B0 benchmark network into five convolutional blocks. The five downsampled convolutional blocks in MA-Net are then sequentially replaced. Furthermore, an extra loss term for the sea-land boundary region is incorporated through the introduction of a boundary region enhancement loss function. This approach encourages the network to focus on learning the boundary region between the sea and land. This improves the accuracy of its prediction. The study presents the results of segmentation experiments conducted on a constructed Gaofen-2 image dataset. The improved EMA-Net model, utilizing the boundary region enhancement loss, achieves better performance than other methods for both the overall region and the sea-land boundary region. The LR (Land Recall), LP (Land Precision), SR (Sea Recall), SP (Sea Precision), F1 Score (F1-Score), mIoU (Mean Intersection over Union), and EA (Edge Accuracy) are averaged over multiple experiments to reach 97.78%, 96.63%, 97.65%, 98.48%, 97.62%, 95.37%, and 87.08% respectively. Additional experiments on the IKONOS images also confirmed the adaptability of the proposed method to high-resolution imagery.