Applied Computing and Geosciences最新文献

Flooding presents a formidable challenge in the United States, endangering lives and causing substantial economic damage, averaging around $5 billion annually. Addressing this issue and improving community resilience is imperative. This project employed machine learning techniques and publicly available data to explore the factors influencing flooding and to develop flood susceptibility maps at various spatial resolutions. Six machine learning algorithms, including Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), K-nearest neighbor (KNN), Adaptive Boosting (Ada Boost), and Extreme Gradient Boosting (XGB) were used. Geospatial datasets comprising thirteen predictor variables and 1528 flood inventory data collected since 1996 were analyzed. The predictor variables are rainfall, elevation, slope, aspect, flow direction, flow accumulation, Topographic Wetness Index (TWI), distance from the nearest stream, evapotranspiration, land cover, impervious surface, land surface temperature, and hydrologic soil group. Five hundred twenty-eight non-flood data points were randomly created using a stream buffer for two scenarios. A total of 2964 data points were classified into flooded (1) and non-flooded (0) categories and used as a target. Overall, testing results showed that the XGB and RF models performed relatively well in both cases over multiple resolutions compared to other models, with an accuracy ranging from 0.82 to 0.97. Variable importance analysis depicted that predictor variables such as distance from the streams, hydrologic soil type, rainfall, elevation, and impervious surfaces significantly affected flood prediction, suggesting a strong association with the underlying driving process. The improved performance and the variation of the susceptible areas across two scenarios showed that considering predictor variables with multiple resolutions and appropriate non-flooding training points is critical for developing flood-susceptibility models. Furthermore, using tree-based ensemble algorithms like RF and XG boost in the stack generalization approach can help achieve robustness in a flood susceptibility model where multiple algorithms are being evaluated.

Floods affect chronically many communities around the world. Their socioeconomic impact increases year-by-year, boosted by global warming and climate change. Combined with long-term preemptive measures, preparatory actions are crucial when floods are imminent. In the last decade, machine learning models have been used to anticipate these hazards. In this work, we model the Ter river (NE Spain), which has historically suffered from floods, using traditional ML models such as K-nearest neighbors, Random forests, XGBoost and Linear regressors. Publicly available river flow and precipitation data was collected from year 2009 to 2021. Our analysis measures the time elapsed between observing a flow rise event at different stations (or heavy rain, for rainfall stations), and use the measured time lags to align the data from the different stations. This information provides increased interpretability to our river flow models and flood forecasters. A thorough evaluation reveals that ML techniques can be used to make short-term predictions of the river flow, even during heavy rain and large flow rise events. Moreover, our flood forecasting system provides valuable interpretable information for setting up timely preparatory actions.

The emission of dust particles, mainly from arid and semi-arid lands, as a result of climate change and human activities, is known to be a global issue. Identifying dust emission sources is the first key step in dealing with the hazardous consequences of this rising phenomenon. This study is an attempt to address one of the major challenges in mapping dust emission sources. Accordingly, an innovative approach based on visual interpretation of multi-temporal MODIS-Terra/Aqua imagery and object-oriented image segmentation techniques has been developed and implemented in the study areas of the Tigris and Euphrates basin and eastern Iran. This approach takes advantage of land surface characteristics (i.e., dust drivers), including geomorphology, soil, land use/cover, and land surface radiation, to attribute dust emission hotspots to their corresponding areas using multi-source remote sensing data. The results show that the multi-resolution segmentation algorithm with optimized parameters can identify homogeneous segments corresponding to dust emission sources in the study areas with an average spatial agreement of ∼92% compared to the reference areas. Our findings emphasize the robustness and generalizability of the proposed approach, and its capabilities can be used in a complementary way with visual interpretation of satellite images to map dust sources with high spatial accuracy.

El Fierro intrusive body is one of the bodies that compose the La Jovita–Las Aguilas mafic–ultramafic belt, located in the Sierra Grande de San Luis, Argentina. The units of this belt carry a base metal sulfide (BMS) mineralization and platinum group minerals (PGM). The macroscopic description of mafic and ultramafic rocks, as is usually done by the mining exploration companies, leads to an imprecise modal classification of the rocks. In this study, we develop a random forest-based prediction model, which uses geochemical parameters to classify mafic and ultramafic rocks intercepted by drill cores. This model showed an accuracy of between 86% and 94%, and an f1_score of 96%. Random forest classification is a widely adopted Machine Learning approach to construct predictive models across various research domains. However, as models become more complex, their interpretation can be considerably difficult. To interpret the model results, we use both global and local perspectives, incorporating the SHAP (SHapley Additive exPlanations) method. The SHAP technique allows us to analyze individual samples using force plots, and provides a measure of the importance of each geochemical input attribute in the model output. As a result of analyzing the contribution of each input feature to the model, the three variables with the highest contributions were identified in the following order: ${\mathrm{Al}}_2{O}_3$, $\mathrm{MgO}$, and $\mathrm{Sr}$.

Carbon-cycle modelling is essential for testing the main carbon sources and sinks as climate forcings, and we introduce and describe GEOCARB_NET, a graphical user interface for the geologic carbon and sulfur cycle model GEOCARBSULFvolc. The software system is menu-driven, user-friendly, and the user is never far removed from the basic input parameters from which atmospheric CO₂ and O₂ concentrations can be derived. GEOCARB_NET is supplied with several published models and the user can easily test and refine these models with different parametrizations. GEOCARB_NET also contains libraries of models and proxy data, which easily can be compared with each other. Our examples focus on how to use GEOCARB_NET in the context of Phanerozoic climate change and highlights how certain key input parameters can seriously affect reconstructed CO₂ levels.

Geostatistical cascade modeling of Mineral Resources is challenging in vein-type gold deposits. The narrow shape and long-range features of these auriferous veins, coupled with the paucity of drill-hole data, can complicate the modeling process and make the use of two-point geostatistical algorithms impractical. Instead, multiple-point geostatistics techniques can be a suitable alternative. However, the most challenging part in implementing the MPS is to use a suitable training data set or training image (TI). In this paper, we suggest using the radial basis function algorithm to build a training image and the DeeSse algorithm, one of the multiple-point statistics (MPS) methods, to model two long-range veins in a gold deposit. It is demonstrated that DeeSse can replicate long-range vein features better than plurigaussian simulation techniques when there is a lack of conditioning data. This is shown by several validation processes, such as comparing simulation results with an interpretive geological block model and replicating geological proportions.

Permeability is a key rock property important for scientific applications that require simulation of fluid flow. Although permeability is determined using core flooding experiments, recent advancements in micro-CT imaging and pore scale fluid flow simulations have made it possible to constrain permeability honoring pore scale rock structure. Previous studies have reported that complex association of pores and solid grains often results in preferential flow paths which influence the resulting velocity field and, hence, the upscaled permeability value. Additionally, the pore structure may change due to geochemical processes such as microbial growth, mineral precipitation and dissolution. This could result in a flow field which dynamically evolves spatially and temporally. It would require numerous experiments or full physics simulations to determine the resultant upscaled Darcy permeability for such dynamically changing systems. This study presents a graph theory-based approach to upscale permeability from pore-to-Darcy scale for changing pore structure. The method involves transforming a given micro-CT rock image to a graph network map followed by the identification of the least resistance path between the inlet and the outlet faces using Dijkstra's algorithm where resistance is quantified as a function of pore sizes. The least resistance path is equivalent to the path of lowest resistance within the domain. The method was tested on micro-CT images of the samples of Sherwood Sandstone, Ketton Limestone and Estaillades Limestone. The three micro-CT images were used to generate 30 synthetic scenarios for geochemically induced pore structure changes covering a range of pore and solid grain growth. The least resistance value obtained from Dijkstra's algorithm was observed to correlate with upscaled permeability value determined from full physics simulations, while improving computational efficiency by a factor of 250. This provides confidence in using graph theory method as a proxy for full physics simulations for determining effective permeability for samples with changing pore structure.

The complex interplay of various complicated meteorological and oceanic processes has made it more difficult to accurately predict Indian monsoon rainfall. A future-oriented and one of the most potential methods for predictive analytics is deep learning. The proposed work exploits empirical Mode Decomposition-Detrended Fluctuation Analysis (EMD-DFA) and long short-term memory (LSTM) deep neural networks (EMD-LSTM) to build novel predictive models and analyze predictability effectively. The time series data of each homogeneous monsoon zone are decomposed into different empirical time series components known as intrinsic mode functions (IMFs). The proposed work's obtained results report that the EMD-LSTM hybrid strategy consistently outperforms other methods in terms of accuracy. Furthermore, we examined possible relationships between each homogeneous monsoon zone and multiple climate drivers, shedding light on the complicated relationships that influence monsoon patterns. This study presents a unique way of predicting complex monsoon rainfall in homogenous regions of India and marks the first application of the EMD-LSTM technique for this purpose to the best of our knowledge which is necessary for improving water conservation and distribution at different climate zones of India.

The complexity of the relationship between climate variables including temperature, precipitation, soil moisture, and the Normalized Difference Vegetation Index (NDVI) arises from the complex interaction between these factors. NDVI is a widely used index to analyze the characteristics of vegetation cover, including its dynamic patterns. It is a crucial parameter for examining vegetation stability, which is vital for ensuring sustainable food production. This study aims to develop a global-scale NDVI forecasting model based on deep learning algorithms that consider climate variables. The model was trained using three years of global data, including NDVI, temperature, precipitation, and soil moisture. The results of this study demonstrate the effectiveness of the deep learning model for forecasting NDVI. The model accurately predicted NDVI values, as evidenced by the high coefficient of determination ($R^2$) values and the negligible average disparity between predicted and observed NDVI values. The study conducted an analysis of the model’s performance both temporally and spatially. The performance of the model was examined for each month and the overall performance of the model for months presented as the model’s temporal performance overall. Additionally, the model’s performance was analyzed at different latitudes, categorized as mid-latitude and low-latitude performance. The temporal analysis of the model demonstrated an overall $R^2$ value of 0.85 and an RMSE of 0.096. Meanwhile, the spatial analysis of the model showed that it performed well at low-latitude, with an $R^2$ value of 0.84 and an RMSE of 0.098, and at mid-latitude, with an $R^2$ value of 0.82 and an RMSE of 0.095. This suggests that the model’s forecasted NDVI values showed a small average difference compared to actual values in both temporal and spatial analyses. Overall, the study supports the idea that deep learning models can effectively forecast NDVI using climate variables across various geographical zones and throughout different months of the year.

This paper documents the development of PixelSWAT, a Graphical User interface (GUI) python toolbox developed with the motive of creating gridded watershed and stream features to run the Soil and Water Assessment Tool (SWAT) in a distributed discretization scheme thus allowing optimum utilization of gridded weather datasets. Additionally, the tool also aims to automate the preparation of SWAT weather input files from Network Common Data (NetCDF) files for any SWAT user along with the option to interpolate the weather files for each grid. A case study was conducted in the Mago basin of Tawang, Arunachal Pradesh, using gridded weather datasets for hydrological simulation. Three SWAT models were prepared – a conventional SWAT model; a 500 m and a 1000 m gridded watershed PixelSWAT models. Statistical indices Nash Sutcliffe (NSE), Coefficient of Determination (R²) and Percent Bias (PBIAS) showed that the PixelSWAT projects performed marginally better than the conventional model and also incorporated the weather data more meaningfully.