Ahmed El-Shafie’s research while affiliated with Al Ain University and other places

In arid and semi-arid regions, flash floods pose significant threats to human life and infrastructure, especially in the geomorphologically intricate environments of wadis. Effective risk management and mitigation require accurate assessment and analysis of flash flood susceptibility. The instrumental role of remote sensing is evident in accurately delineating land use and land cover (LULC), a foundational layer for flash flood risk assessment. The primary phase of this study is to evaluate the quality of the LULC classification using Sentinel-2 and Landsat-8 satellite imagery. The Gharbia Governorate in Egypt, with its varied land uses including Built Area, Water, Crops, and Agricultural Land, was exploited for this assessment. With the integration of remote sensing, GIS, and MCDM methods, this study aims to enhance the reliability of flood risk evaluations. Morphometric factors such as soil, geology, drainage, slope, LULC, and elevation were assessed, weighted, and integrated using the Analytic Hierarchy Process (AHP) to delineate flash flood risk zones in Wadi El-Assiuty, Egypt. Findings indicate that Sentinel-2 data outperforms Landsat-8 in LULC classification accuracy. The flash flood risk map reveals that most areas in Wadi El-Assiuty exhibit moderate to high inundation risk. The weight results were consistent with previous studies, while the drainage is identified as the most significant factor influencing flood risk (28%), followed by geology (23%) and LULC (16%). Furthermore, the case study risk map reveals that the southwestern region is identified as a high-risk area due to its high drainage capacity, low elevation, extensive urbanization, and geological features like fanglomerate and Pliocene deposits. These insights are valuable for land-use planning and disaster preparedness in arid regions.

Accurately classifying river inflow is crucial for understanding river dynamics and ecosystem health. This study evaluates the performance of seven machine learning models, including Support Vector Machines (SVM), Logistic Regression (LR), K-Nearest Neighbors (KNN), Decision Trees (DT), Random Forests (RF), Adaptive Boosting (AdaBoost), and Multi-Layer Perceptron (MLP), for streamflow classification. One of the key challenges in this task is the imbalance in class distributions, which can negatively impact model performance. To address this, we apply the Synthetic Minority Over-sampling Technique (SMOTE) to improve classification outcomes for minority classes. Furthermore, we investigate the impact of four proposed feature selection methods, including mutual information (MI-FS), linear kernel SVM (SVM-FS), random forest (RF-FS), and multi-criteria selection (MC-FS) on model performance by identifying optimal lag values. Model hyperparameters are fine-tuned using GridSearchCV technique, and evaluation step is assessed across seven performance metrics. Experimental results show that MLP and SVM consistently outperform other models, making them the most suitable choices for streamflow classification. Among the FS techniques, MC-FS demonstrates superior performance by effectively reducing dimensionality while preserving predictive power. However, our findings indicate that SMOTE enhances classification for minority classes but reduces accuracy for majority classes, highlighting a trade-off in handling imbalanced data. Additionally, we observe that the linear assumption in SVM-FS can negatively impact model performance when it fails to detect all relevant input features. These insights provide valuable guidance for future streamflow classification tasks.

Groundwater (GW) acts as a key source of freshwater for household, industrial, and agricultural use across the globe. Assessing groundwater quality is important, given the increasing pressures from human activities and climate change. However, collecting groundwater quality data is challenging due to high costs and time-intensive processes. Recently, integrating the Groundwater Quality Index (GWQI) and machine learning (ML) has emerged as a promising approach for managing groundwater quality, though advancements specifically addressing time and cost-efficiency remain limited. The study analysed groundwater data from Erbil Basin, Kurdistan, Iraq, using 13 quality-impacting parameters, resulting in over 66,000 data points. The Weighted Arithmetic Water Quality Index (WAWQI) method provided a comprehensive evaluation by integrating multiple parameters into a single quantitative assessment. Machine learning models, including Artificial Neural Networks (ANN), Gaussian Process Regression (GPR), ensemble methods, and Support Vector Machines (SVM), were employed to enhance prediction accuracy. Seven scenarios explored the effects of excluding specific parameters, such as chloride (Cl⁻), nitrate (NO₃⁻), and sulfate (SO₄²⁻). Results showed exponential GPR performed best in scenarios 1–1 (96.10%), 2–3 (96.32%), and 3 − 1 (93.9%). Linear SVM achieved the highest accuracy in scenarios 1–2 (95.58%) and 2 − 1 (93.24%), while the wide neural network excelled with perfect accuracy in scenario 2–2. Scenario 1–3’s top performance was by exponential GPR with 93.29% accuracy. These findings highlight the potential of ML models in optimizing groundwater quality predictions while addressing cost and time constraints.

Towards a better groundwater management, developing a prediction model for groundwater quality is of utmost importance. The conventional method of measuring groundwater quality data often associated with errors due to the lengthy duration of investigation of the parameters as well as the tremendous effort and time involved in gathering and analysing the samples. The expense associated with determining the parameters’ values via laboratory testing is substantial. There has been a notable increase in machine learning (ML) application for modelling groundwater quality as of recent, evidenced by a wealth of studies reporting impressive results. This paper provides an extensive examination of 91 relevant articles picked from the Web of Science and PubMed, from 2015 to 2024. The focus of the review revolves on significant ML algorithms, including artificial neural networks (ANN), random forest (RF), support vector machines (SVM), hybrid models, and other algorithms that have demonstrated efficacy in predicting groundwater quality, such as k-nearest neighbours and extreme gradient boosting (XGBoost). Critical modelling concepts such as data splitting, utilized parameters, performance metrics, and study areas were addressed, emphasizing optimal practices for effective groundwater quality prediction with ML.

Foam glass (FG) is characterized by its lightweight structure and exceptional insulating properties, making it a highly suitable material for a wide range of applications. It is produced by foaming molten glass, resulting in a cellular structure that enhances its insulation and impact-resistant properties. Due to its sustainability and durability, FG is increasingly used in the construction, automotive, and packaging sectors. In this context, the current study proposes a novel approach by developing a thoughtful system for assessing performance and intelligent design utilizing ML models such as Gradient Boosting (GB), Random Forest (RF), Gaussian Process Regression (GPR), and Linear Regression (LR) to predict porosity and compressive strength (CS) of FG. The dataset comprises 214 data points, encompassing input variables such as glass particle diameter, foam agent content, heating rate, holding time, sintering temperature, and dry density, with output parameters of porosity and CS. Data preprocessing involved Pearson correlation analysis to address multicollinearity and reveal nonlinear relationships among variables. Model performance was evaluated through R-values, mean absolute error, and root mean square error metrics, demonstrating that the GPR model achieved superior prediction accuracy with R-values of 0.91 and 0.82 for porosity and CS, respectively. The GB model followed closely, while the RF and LR models showed lower accuracy. Partial dependence plots and global feature importance analyses highlighted density and foam agent content as critical factors influencing FG properties. By achieving the most precise predictions with minimal error distributions, the GPR model offers actionable insights into FG design. These findings enable the optimization of FG production by providing reliable tools for predicting and controlling porosity and CS, reducing material waste, enhancing product quality, and streamlining manufacturing processes. This study demonstrates the potential of advanced ML techniques to bridge the gap between predictive modeling and practical applications in the digital design of FG.