Earth Science Informatics最新文献

In recent years, the effects and consequences of climate change have shown themselves by creating irregularities and trends in the essential climatic variables. In most cases, the trend of climatic variables is associated with periodicity. In this study, the trends and periodicity of these data (precipitation, temperature, evapotranspiration, and net available water (NWA) have been investigated in a period of 60 years in Iran. The Mann–Kendall trend test and Sen’s slope estimator are applied for analyzing the trend and its magnitude. Wavelet transform is used to detect the periodicity of time series and to determine the correlation between NWA and temperature, precipitation, and evapotranspiration in common periodicity. The results show that the stations located in eastern and western Iran have more significant increasing/decreasing trends. Evapotranspiration shows the highest increasing trend in most stations, followed by temperature, while NWA and precipitation have trends at lower significance levels and decreasing direction. The examination of periodicity in time series showed that, among all the studied stations, evapotranspiration has the most extended periodicity with an average length of 8.3 years, followed by NWA, temperature, and precipitation with 7.3 years, 5.8 years, and 5.5 years. The results of the correlations investigation showed that in about 80% of the stations, there is a high correlation between precipitation and NWA in the short-term periodicity and at the end of the studied period. The evapotranspiration variable in most stations has a high correlation in different periodicities with the amount of NWA.

With the rapid development of remote sensing technology and the widespread application of remote sensing images, remote sensing object detection has become a hot research direction. However, we observe three primary challenges in remote sensing object detection: scale variations, small objects, and complex backgrounds. To address these challenges, we propose a novel detector, he Multi-Scale Context-Aware Network (MSCANet). First, we introduce a Multi-Scale Fusion Module (MSFM) that provides various scales of receptive fields to extract contextual information of objects at different scales adequately. Second, the Multi-Scale Guidance Module (MSGM) is proposed, which fuses deep and shallow feature maps from multiple scales, reducing the loss of feature information in small objects. Finally, we introduce the Context-Aware DownSampling Module (CADM). It dynamically adjusts context information weights at different scales, effectively reducing interference from complex backgrounds. Experimental results demonstrate that the proposed MSCANet achieves superior performance results with mean average precision (mAP) of 97.1% and 73.4% on the challenging RSOD and DIOR datasets, respectively, which indicates that the proposed network is suitable for remote sensing object detection and is of a great reference value.

Soil organic carbon (SOC) is the main component of soil organic matter (SOM) and constitutes the crucial component of the soil. It supports key soil functions, stabilizes soil structure, aid in plant-nutrient retention and release, and promote water infiltration and storage. Predicting SOC using Sentinel-2 data integrated with machine learning algorithms under zero tillage practice is inadequately documented for developing countries like Zimbabwe. The purpose of this study is to evaluate the performance of support vector machine (SVM), artificial neural network (ANN), and partial least square regression (PLSR) algorithms from Sentinel-2 data for SOC estimation. The SVM, ANN and PLSR models were used with a cross-validation to estimate the SOC content based on 50 georeferenced calibration samples under a zero-tillage practice. The ANN model outperformed the other two models by delivering a coefficient of determination (R²) of between 55 and 60% of SOC variability and RMSE varied between 5.01 and 8.78%, whereas for the SVM, R² varied between 0.53 and 0.57 and RMSE varied between 6.25 and 11.39%. The least estimates of SOC provided by the PLSR algorithm were, R² = 0.44–0.49 and RMSE = 7.59–12.42% for the top 15 cm depth. Results with and R², root mean square error (RMSE) and mean absolute error (MAE) for SVM, ANN and PLSR, show that the ANN model is highly capable for capturing SOC variability. Although the ANN algorithm provides more accurate SOC estimates than the SVM algorithm, the difference in accuracy is not significant. Results revealed a satisfactory agreement between the SOC content and zero tillage practice (R², coefficient of variation (CV), MAE, and RMSE using SVM, ANN and PLSR for the validation dataset using four predictor variables. The calibration results of SOC indicated that the mean SOC was 15.83% and the validation mean SOC was 17.02%. The SOC validation dataset (34.17%) had higher degree of variation around its mean as compared to the calibration dataset (29.86%). The SOC prediction results can be used as an important tool for informed decisions about soil health and productivity by the farmers, land managers and policy makers.

Rapid identification and detection of landslides is of significance for disaster damage assessment and post-disaster relief. However, U-net for rapid landslide identification and detection suffers from semantic gap and loss of spatial information. For this purpose, this paper proposed the U-net with a progressive Convolutional Block Attention Module (CBAM-U-net) for landslide boundary identification and extraction from high-precision aerial imagery. Firstly, 109 high-precision aerial landslide images were collected, and the original database was extended by data enhancement to strengthen generalization ability of models. Subsequently, the CBAM-U-net was constructed by introducing spatial attention module and channel attention module for each down-sampling process in U-net. Meanwhile, U-net, FCN and DeepLabv3 + are used as comparison models. Finally, 6 evaluation metrics were used to comprehensively assess the ability of models for landslide identification and segmentation. The results show that CBAM-U-net exhibited better recognition and segmentation accuracies compared to other models, with optimal values of average row correct, dice coefficient, global correct, IoU and mean IoU of 98.3, 0.877, 95, 88.5 and 90.2, respectively. U-net, DeepLab V3 + , and FCN tend to confuse bare ground and roads with landslides. In contrast, CBAM-U-net has stronger ability of feature learning, feature representation, feature refinement and adaptation.The proposed method can improve the problems of semantic gap and spatial information loss in U-net, and has better accuracy and robustness in recognizing and segmenting high-precision landslide images, which can provide certain reference value for the research of rapid landslide recognition and detection.

At a particular location on the ground, geotechnical measurements of soil properties are utilized to offer information for infrastructure design. Design uncertainty and dependability may increase when little point data is used. Geophysical techniques offer constant geographic information about the soil and are less time-consuming and intrusive. Geophysical data, however, is not expressed in terms of technical specifications. To enable the use of geophysical data in geotechnical designs, correlations between geotechnical and geophysical characteristics are required. The S- and P- seismic wave velocities are the main focus of the present geophysical technique research. Artificial neural network (ANN) models are developed using published data to predict seismic wave velocity and soil classification for seismic site effect evaluation. The results of ANN models using publicly available data demonstrate that seismic wave velocity has a moderate to high degree of accuracy in predicting soil classification. Regression is not as effective as artificial neural networks (ANN) in terms of overall performance. To confirm this, enclosed areas were evaluated to accurately predict soil classification and assess the performance of both ANN and regression models. The artificial neural network predicted the enclosed areas with much higher accuracy.

Three supervised machine learning (ML) classification algorithms: Support Vector Classifier (SVC), K- Nearest Neighbour (K-NN), and Linear Discriminant Analysis (LDA) classification algorithms are combined with seventy-six (76) data points of nine (9) core sample datasets retrieved from five (5) selected wells in oilfields of the Subei Basin to delineate bioturbation. Application of feature selection via p-score and f-scoring reduced the number of relevant features to 7 out of the 12 considered. Each classifier underwent model training and testing allocating 80% of the data for training and the remaining 20% for testing. Under the model training, optimization of hyperparameters of the SVC (C, Gamma and Kernel) and K-NN (K value) was performed via the grid search to understand the best form of the decision boundaries that provides optimal accuracy of prediction of Bioturbation. Results aided the selection of optimized SVC hyperparameters such as a linear kernel, C-1000 and Gamma parameter—0.10 that provided a training accuracy of 96.17%. The optimized KNN classifier was obtained based on the K = 5 nearest neighbour to obtain a training accuracy of 73.28%. The training accuracy of the LDA classifier was 67.36% which made it the worst-performing classifier in this work. Further cross-validation based on a fivefold stratification was performed on each classifier to ascertain model generalization and stability for the prediction of unseen test data. Results of the test performance of each classifier indicated that the SVC was the best predictor of the bioturbation index at 92.86% accuracy, followed by the K-NN model at 90.48%, and then the LDA classifier which gave the lowest test accuracy at 76.2%. The results of this work indicate that bioturbation can be predicted via ML methods which is a more efficient and effective means of rock characterization compared to conventional methods used in the oil and gas industry.

Susceptibility mapping has been an effective approach to manage the threat of debris flows. However, the sample heterogeneity problem has rarely been considered in previous studies. This paper is to explore the effect of sample heterogeneity on susceptibility mapping and propose corresponding solutions. Two unsupervised clustering approaches including K-means clustering and fuzzy C-means clustering were introduced to divide the study area into several homogeneous regions, each region was processed independently to solve the sample heterogeneity problem. The information gain ratio method was used to evaluate the predictive ability of the conditioning factors in the total dataset before clustering and the homogeneous datasets after clustering. Then the total dataset and the homogeneous datasets were involved in the random forest modeling. The receiver operating characteristic curves and related statistical results were employed to evaluate the model performance. The results showed that there was a significant sample heterogeneity problem for the study area, and the fuzzy C-means algorithm can play an important role in solving this problem. By dividing the study area into several homogeneous regions to process independently, conditioning factors with better predictive ability, models with better performance and debris flow susceptibility maps with higher quality could be obtained.

Hyperspectral Images (HSIs) possess extensive applications in remote sensing, especially material discrimination and earth observation monitoring. However, constraints in spatial resolution increase sensitivity to spectral noise, limiting the ability to adjust Receptive Fields (RFs). Convolutional Neural Networks (CNNs) with fixed RFs are a common choice for HSI classification tasks. However, their potential in leveraging the appropriate RF remains under-exploited, thus affecting feature discriminative capabilities. This study introduces an Enhanced Adaptive Source-Selection Kernel with Attention Mechanism (EAS(^2)KAM) for HSI Classification. The model incorporates a Three Dimensional Enhanced Function Mixture (3D-EFM) with a distinct RF for local low-rank contextual exploitation. Furthermore, it incorporates diverse global RF branches enriched with spectral attention and an additional spectral-spatial mixing branch to adjust RFs, enhancing multiscale feature discrimination. The 3D-EFM is integrated with a 3D Residual Network (3D ResNet) that includes a Channel-Pixel Attention Module (CPAM) in each segment, improving spectral-spatial feature utilization. Comprehensive experiments on four benchmark datasets show marked advancements, including a maximum rise of 0.67% in Overall Accuracy (OA), 0.87% in Average Accuracy (AA), and 1.33% in the Kappa Coefficient ((kappa )), outperforming the top two HSI classifiers from a list of eleven state-of-the-art deep learning models. A detailed ablation study evaluates model complexity and runtime, confirming the superior performance of the proposed model.

Automated methods for building function classification are essential due to restricted access to official building use data. Existing approaches utilize traditional Natural Language Processing (NLP) techniques to analyze textual data representing human activities, but they struggle with the ambiguity of semantic contexts. In contrast, Large Language Models (LLMs) excel at capturing the broader context of language. This study presents a method that uses LLMs to interpret OpenStreetMap (OSM) tags, combining them with physical and spatial metrics to classify urban building functions. We employed an XGBoost model trained on 32 features from six city datasets to classify urban building functions, demonstrating varying F1 scores from 67.80% in Madrid to 91.59% in Liberec. Integrating LLM embeddings enhanced the model's performance by an average of 12.5% across all cities compared to models using only physical and spatial metrics. Moreover, integrating LLM embeddings improved the model's performance by 6.2% over models that incorporate OSM tags as one-hot encodings, and when predicting based solely on OSM tags, the LLM approach outperforms traditional NLP methods in 5 out of 6 cities. These results suggest that deep contextual understanding, as captured by LLM embeddings more effectively than traditional NLP approaches, is beneficial for classification. Finally, a Pearson correlation coefficient of approximately -0.858 between population density and F1-scores suggests that denser areas present greater classification challenges. Moving forward, we recommend investigation into discrepancies in model performance across and within cities, aiming to identify generalized models.

Many landslides occurred every year, causing extensive property losses and casualties in China. Landslide susceptibility mapping is crucial for disaster prevention by the government or related organizations to protect people's lives and property. This study compared the performance of random forest (RF), classification and regression trees (CART), Bayesian network (BN), and logistic model trees (LMT) methods in generating landslide susceptibility maps in Yanchuan County using optimization strategy. A field survey was conducted to map 311 landslides. The dataset was divided into a training dataset and a validation dataset with a ratio of 7:3. Sixteen factors influencing landslides were identified based on a geological survey of the study area, including elevation, plan curvature, profile curvature, slope aspect, slope angle, slope length, topographic position index (TPI), terrain ruggedness index (TRI), convergence index, normalized difference vegetation index (NDVI), distance to roads, distance to rivers, rainfall, soil type, lithology, and land use. The training dataset was used to train the models in Weka software, and landslide susceptibility maps were generated in GIS software. The performance of the four models was evaluated by receiver operating characteristic (ROC) curves, confusion matrix, chi-square test, and other statistical analysis methods. The comparison results show that all four machine learning models are suitable for evaluating landslide susceptibility in the study area. The performances of the RF and LMT methods are more stable than those of the other two models; thus, they are suitable for landslide susceptibility mapping.