Conventional methods of estimating pressure coefficients of buildings retain time and cost constraints. Recently, machine learning (ML) has been successfully established to predict wind pressure coefficients. However, regardless of the accuracy, ML models are incompetent in providing end-users’ confidence as a result of the black-box nature of predictions. In this study, we employed tree-based regression models (Decision Tree, XGBoost, Extra-tree, LightGBM) to predict surface-averaged mean pressure coefficient (C_p,mean), fluctuation pressure coefficient (C_p,rms), and peak pressure coefficient (C_p,peak) of low-rise gable-roofed buildings. The accuracy of models was verified using Tokyo Polytechnic University (TPU) wind tunnel data. Subsequently, we used Shapley Additive Explanations (SHAP) to explain the black-box nature of the ML predictions. The comparison revealed that tree-based models are efficient and accurate in wind-predicting pressure coefficients. Interestingly, SHAP provided human-comprehensible explanations for the interaction of variables, the importance of features towards the outcome, and the underlying reasoning behind the predictions. Moreover, SHAP confirmed that tree-based predictions adhere to the flow physics of wind engineering, advancing the fidelity of ML-based predictions.

Predicting the bulk-average velocity (U_B) in open channels with rigid vegetation is complicated due to the non-linear nature of the parameters. Despite their higher accuracy, existing regression models fail to highlight the feature importance or causality of the respective predictions. Therefore, we propose a method to predict U_B and the friction factor in the surface layer (f_S) using tree-based machine learning (ML) models (decision tree, extra tree, and XGBoost). Further, Shapley Additive exPlanation (SHAP) was used to interpret the ML predictions. The comparison emphasized that the XGBoost model is superior in predicting U_B (R = 0.984) and f_S (R = 0.92) relative to the existing regression models. SHAP revealed the underlying reasoning behind predictions, the dependence of predictions, and feature importance. Interestingly, SHAP adheres to what is generally observed in complex flow behavior, thus, improving trust in predictions.

Traditionally, ASR expansion is determined using experimental and finite element method (FEM) based numerical modelling. However, these methods are time-consuming and computationally costly, which makes ASR prediction challenging. Machine learning (ML) techniques can serve as effective alternatives for the early detection of expansion in concrete structures. In this study, we developed four different machine learning models – extreme gradient boosting (XGBoost), random forest (RF), support vector regression (SVR), and k-nearest neighbor (KNN) to predict the ASR expansion in concrete using a comprehensive dataset with 1896 data points. Each model was evaluated based on the model performance and XGBoost shows the most effective model for predicting the ASR expansion with R² of 0.99 for training and R² of 0.98 for testing. After identifying the best-performing model, Shapley Additive Explanations (SHAP) were employed to interpret its predictions. This approach provides insights into the model’s decision-making process, clarifying the complex nature of machine learning algorithms. Using Shapley Additive Explanations (SHAP), time and the average size of reactive aggregate were identified as critical parameters influencing ASR expansion. This research holds significant value for the construction industry, as accurately predicting ASR expansion can lead to optimized material usage and improved structural performance.

Storms, including cyclones, hurricanes, and typhoons, pose significant threats due to their destructive winds and heavy rainfall, causing extensive damage to infrastructure, ecosystems, and human lives. Accurate prediction of storm paths and intensities is crucial for disaster preparedness and minimizing losses. Despite advances in meteorological science and computational modeling, accurately forecasting sudden changes in storm behavior, such as rapid intensification or unexpected weakening, remains challenging. Traditional numerical weather prediction (NWP) models integrate physical laws governing atmospheric conditions but struggle with uncertainties stemming from rapid weather variations, incomplete observations, and computational constraints. This study proposes a hybrid approach integrating sequence-based deep learning (Long Short-Term Memory (LSTM) and Transformers) with gradient boosting techniques (XGBoost, LightGBM) for predicting storm tracks and intensities. Additionally, physics-informed constraints such as storm kinetic energy, Coriolis force, and land interaction effects are incorporated to enhance the physical consistency and interpretability of the models. By combining machine learning models with fundamental meteorological principles, this work aims to improve the reliability and accuracy of cyclone forecasts. The main contributions of this study include: (1) the development of a multi-modal hybrid model combining deep learning and tree-based machine learning methods; (2) incorporation of physics-informed constraints to enhance model generalization and physical validity; and (3) rigorous evaluation of model performance using real-world cyclone datasets with robust metrics, including the Haversine distance for path prediction and Mean Absolute Error (MAE) for intensity estimation.

Studying future precipitation behaviour in river basins is essential for proper water resources and land-use planning within them, as this will help to reduce the risk and mitigate disasters that can occur in the future. General Circulation Models (GCMs) are used to study future precipitation fluctuations, which simulate large-scale climate variations under the effect of greenhouse gas changes. The GCM runs at a coarse spatial resolution which cannot be directly used for climate impact studies. Therefore, downscaling is required to extract the sub-grid and local scale information. This study examines the use of the Long Short-Term Memory (LSTM) neural network for climate downscaling to the Mi-Oya river basin in Sri Lanka using CNRM-CM5 and HadCM3 GCMs and observed annual data for 35 years. The precipitation data were extracted to cover Sri Lanka. Current downscaling models mostly use Convolutional Neural Networks (CNNs) to downscale GCMs. Out of 42 GCMs, two appropriate GCMs were chosen using the data analysis tool Data Integration and Analysis System (DIAS). The best predictor variables were chosen using the LASSO regression method. In this research, Machine Learning models were implemented using the Google TensorFlow platform. The Nash–Sutcliffe coefficient, Pearson correlation coefficient, and root-mean-square error performance indices were used to evaluate the performances of different downscaling models. Statistical downscaling was performed on the data at RCP 2.6, 4.5, and 8.5 using a LSTM. Subsequently, the changes that would take place by the year 2100 were analysed. The results show that precipitation will be reduced in the 2nd and 3rd decades of the 21st century, and precipitation will increase toward the 22nd century.