Multi-step ahead forecasting of electrical conductivity in rivers by using a hybrid Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) model enhanced by Boruta-XGBoost feature selection algorithm.

Electrical conductivity (EC) is widely recognized as one of the most essential water quality metrics for predicting salinity and mineralization. In the current research, the EC of two Australian rivers (Albert River and Barratta Creek) was forecasted for up to 10 days using a novel deep learning algorithm (Convolutional Neural Network combined with Long Short-Term Memory Model, CNN-LSTM). The Boruta-XGBoost feature selection method was used to determine the significant inputs (time series lagged data) to the model. To compare the performance of Boruta-XGB-CNN-LSTM models, three machine learning approaches-multi-layer perceptron neural network (MLP), K-nearest neighbour (KNN), and extreme gradient boosting (XGBoost) were used. Different statistical metrics, such as correlation coefficient (R), root mean square error (RMSE), and mean absolute percentage error, were used to assess the models' performance. From 10 years of data in both rivers, 7 years (2012-2018) were used as a training set, and 3 years (2019-2021) were used for testing the models. Application of the Boruta-XGB-CNN-LSTM model in forecasting one day ahead of EC showed that in both stations, Boruta-XGB-CNN-LSTM can forecast the EC parameter better than other machine learning models for the test dataset (R = 0.9429, RMSE = 45.6896, MAPE = 5.9749 for Albert River, and R = 0.9215, RMSE = 43.8315, MAPE = 7.6029 for Barratta Creek). Considering the better performance of the Boruta-XGB-CNN-LSTM model in both rivers, this model was used to forecast 3-10 days ahead of EC. The results showed that the Boruta-XGB-CNN-LSTM model is very capable of forecasting the EC for the next 10 days. The results showed that by increasing the forecasting horizon from 3 to 10 days, the performance of the Boruta-XGB-CNN-LSTM model slightly decreased. The results of this study show that the Boruta-XGB-CNN-LSTM model can be used as a good soft computing method for accurately predicting how the EC will change in rivers.

Comparative study for coastal aquifer vulnerability assessment using deep learning and metaheuristic algorithms.

Coastal aquifer vulnerability assessment (CAVA) studies are essential for mitigating the effects of seawater intrusion (SWI) worldwide. In this research, the vulnerability of the coastal aquifer in the Lahijan region of northwest Iran was investigated. A vulnerability map (VM) was created applying hydrogeological parameters derived from the original GALDIT model (OGM). The significance of OGM parameters was assessed using the mean decrease accuracy (MDA) method, with the current state of SWI emerging as the most crucial factor for evaluating vulnerability. To optimize GALDIT weights, we introduced the biogeography-based optimization (BBO) and gray wolf optimization (GWO) techniques to obtain to hybrid OGM-BBO and OGM-GWO models, respectively. Despite considerable research focused on enhancing CAVA models, efforts to modify the weights and rates of OGM parameters by incorporating deep learning algorithms remain scarce. Hence, a convolutional neural network (CNN) algorithm was applied to produce the VM. The area under the receiver-operating characteristic curves for OGM-BBO, OGM-GWO, and VMCNN were 0.794, 0.835, and 0.982, respectively. According to the CNN-based VM, 41% of the aquifer displayed very high and high vulnerability to SWI, concentrated primarily along the coastline. Additionally, 32% of the aquifer exhibited very low and low vulnerability to SWI, predominantly in the southern and southwestern regions. The proposed model can be extended to evaluate the vulnerability of various coastal aquifers to SWI, thereby assisting land use planers and policymakers in identifying at-risk areas. Moreover, deep-learning-based approaches can help clarify the associations between aquifer vulnerability and contamination resulting from SWI.

Prediction of evapotranspiration variance in the Budyko framework with the incorporation of soil storage and runoff.

Water availability needs to be accurately assessed to understand and effectively manage hydrologic environments. However, the estimation of evapotranspiration (ET) is prone to errors due to the complex interactions that occur between the atmosphere, the Earth's surface, and vegetation cover. This paper proposes a novel approach for analyzing the sources of inaccuracy in estimating the annual ET using the Budyko framework (BF), particularly temporal variability in precipitation (P), potential evapotranspiration (EP), runoff (R), and the change in soil storage (ΔS). Error decomposition is employed to determine the individual contributions of P, R, EP, and ΔS to the ET error variance at 12 locations in the state of Illinois using a dataset covering a 22-year period. To the best of our knowledge, this study represents the first BF-based investigation that considers R in the error decomposition of the predicted ET variance. The ET error variance increases with the variance in the P and R in Illinois and decreases with the covariance between these two variables. In addition, when accounting for ΔS in the BF, the scenario in which ΔS affects the total available water (i.e., P) is reliable, with a low prediction error and a 13.87 % lower root mean square error compared with the scenario in which the effect of ΔS is negligible. We thus recommend the inclusion of ΔS and R as key variables in the BF to improve water budget estimations.

Improving the prediction of wildfire susceptibility on Hawai'i Island, Hawai'i, using explainable hybrid machine learning models.

This study presents a comparative analysis of four Machine Learning (ML) models used to map wildfire susceptibility on Hawai'i Island, Hawai'i. Extreme Gradient Boosting (XGBoost) combined with three meta-heuristic algorithms - Whale Optimization (WOA), Black Widow Optimization (BWO), and Butterfly Optimization (BOA) - were employed to map areas susceptible to wildfire. To generate a wildfire inventory, 1408 wildfire points were identified within the study area from 2004 to 2022. The four ML models (XGBoost, WOA-XGBoost, BWO-XGBoost, and BOA-XGBoost) were run using 14 wildfire-conditioning factors categorized into four main groups: topographical, meteorological, vegetation, and anthropogenic. Six performance metrics - sensitivity, specificity, positive predictive values, negative predictive values, the Area Under the receiver operating characteristic Curve (AUC), and the average precision (AP) of Precision-Recall Curves (PRCs) - were used to compare the predictive performance of the ML models. The SHapley Additive exPlanations (SHAP) framework was also used to interpret the importance values of the 14 influential variables for the modeling of wildfire on Hawai'i Island using the four models. The results of the wildfire modeling indicated that all four models performed well, with the BWO-XGBoost model exhibiting a slightly higher prediction performance (AUC = 0.9269), followed by WOA-XGBoost (AUC = 0.9253), BOA-XGBoost (AUC = 0.9232), and XGBoost (AUC = 0.9164). SHAP analysis revealed that the distance from a road, annual temperature, and elevation were the most influential factors. The wildfire susceptibility maps generated in this study can be used by local authorities for wildfire management and fire suppression activity.

Enhancing flood susceptibility modeling using multi-temporal SAR images, CHIRPS data, and hybrid machine learning algorithms.

Flood susceptibility maps are useful tool for planners and emergency management professionals in the early warning and mitigation stages of floods. In this study, Sentinel-1 dB radar images, which provide Synthetic-Aperture Radar (SAR) data were used to delineate flooded and non-flooded locations. 12 input parameters, including elevation, lithology, drainage density, rainfall, Normalized Difference Vegetation Index (NDVI), curvature, ground slope, Stream Power Index (SPI), Topographic Wetness Index (TWI), soil, Land Use Land Cover (LULC), and distance from the river, were selected for model development. The importance of each input parameter on flood occurrences was assessed via the Mutual Information (MI) technique. Several machine learning models, including Radial Basis Function (RBF), and three hybrid models of Bagging (BA-RBF), Random Committee (RC-RBF), and Random Subspace (RSS-RBF), were developed to delineate flood susceptibility areas at Goorganrood watershed, Iran. The performance of each model was evaluated using several error indicators, including correlation coefficient (r), Nash Sutcliffe Efficiency (NSE), Mean Absolute Error (MSE), Root Mean Square Error (RMSE), and the Area Under the Receiver Operating Characteristic (ROC) Curve (AUC). The results showed that the hybrid techniques enhanced the modeling performance of the standalone model, and generally, all hybrid models are more accurate than the standalone model. Although all developed models have performed well, RC-RBF outperforms all of them (AUC = 0.997), followed by BA-RBF (AUC = 0.996), RSS-RBF (AUC = 0.992), and RBF (AUC = 0.975). Generally, about 12 % of the study area has high and very high susceptibility to future flood occurrences.

BattleSound: A Game Sound Benchmark for the Sound-Specific Feedback Generation in a Battle Game.

A haptic sensor coupled to a gamepad or headset is frequently used to enhance the sense of immersion for game players. However, providing haptic feedback for appropriate sound effects involves specialized audio engineering techniques to identify target sounds that vary according to the game. We propose a deep learning-based method for sound event detection (SED) to determine the optimal timing of haptic feedback in extremely noisy environments. To accomplish this, we introduce the BattleSound dataset, which contains a large volume of game sound recordings of game effects and other distracting sounds, including voice chats from a PlayerUnknown's Battlegrounds (PUBG) game. Given the highly noisy and distracting nature of war-game environments, we set the annotation interval to 0.5 s, which is significantly shorter than the existing benchmarks for SED, to increase the likelihood that the annotated label contains sound from a single source. As a baseline, we adopt mobile-sized deep learning models to perform two tasks: weapon sound event detection (WSED) and voice chat activity detection (VCAD). The accuracy of the models trained on BattleSound was greater than 90% for both tasks; thus, BattleSound enables real-time game sound recognition in noisy environments via deep learning. In addition, we demonstrated that performance degraded significantly when the annotation interval was greater than 0.5 s, indicating that the BattleSound with short annotation intervals is advantageous for SED applications that demand real-time inferences.

Prediction of monthly dry days with machine learning algorithms: a case study in Northern Bangladesh.

Dry days at varied scale are an important topic in climate discussions. Prolonged dry days define a dry period. Dry days with a specific rainfall threshold may visualize a climate scenario of a locality. The variation of monthly dry days from station to station could be correlated with several climatic factors. This study suggests a novel approach for predicting monthly dry days (MDD) of six target stations using different machine learning (ML) algorithms in Bangladesh. Several rainfall thresholds were used to prepare the datasets of monthly dry days (MDD) and monthly wet days (MWD). A group of ML algorithms, like Bagged Trees (BT), Exponential Gaussian Process Regression (EGPR), Matern Gaussian Process Regression (MGPR), Linear Support Vector Machine (LSVM), Fine Trees (FT) and Linear Regression (LR) were evaluated on building a competitive prediction model of MDD. In validation of the study, EGPR-based models were able to better capture the monthly dry days (MDD) over Bangladesh compared to those by MGPR, LSVM, BT, LR and FT-based models. When MDD were the predictors for all six target stations, EGPR produced highest mean R2 of 0.91 (min. 0.89 and max. 0.92) with a least mean RMSE of 2.14 (min. 1.78 and max. 2.69) compared to other models. An explicit evaluation of the ML algorithms using one-year lead time approach demonstrated that BT and EGPR were the most result-oriented algorithms (R2 = 0.78 for both models). However, having a least RMSE, EGPR was chosen as the best model in one year lead time. The dataset of monthly dry-wet days was the best predictor in the lead-time approach. In addition, sensitivity analysis demonstrated sensitivity of each station on the prediction of MDD of target stations. Monte Carlo simulation was introduced to assess the robustness of the developed models. EGPR model declared its robustness up to certain limit of randomness on the testing data. The output of this study can be referred to the agricultural sector to mitigate the impacts of dry spells on agriculture.

Uncertainty quantification of granular computing-neural network model for prediction of pollutant longitudinal dispersion coefficient in aquatic streams.

Discharge of pollution loads into natural water systems remains a global challenge that threatens water and food supply, as well as endangering ecosystem services. Natural rehabilitation of contaminated streams is mainly influenced by the longitudinal dispersion coefficient, or the rate of longitudinal dispersion (Dx), a key parameter with large spatiotemporal fluctuations that characterizes pollution transport. The large uncertainty in estimation of Dx in streams limits the water quality assessment in natural streams and design of water quality enhancement strategies. This study develops an artificial intelligence-based predictive model, coupling granular computing and neural network models (GrC-ANN) to provide robust estimation of Dx and its uncertainty for a range of flow-geometric conditions with high spatiotemporal variability. Uncertainty analysis of Dx estimated from the proposed GrC-ANN model was performed by alteration of the training data used to tune the model. Modified bootstrap method was employed to generate different training patterns through resampling from a global database of tracer experiments in streams with 503 datapoints. Comparison between the Dx values estimated by GrC-ANN to those determined from tracer measurements shows the appropriateness and robustness of the proposed method in determining the rate of longitudinal dispersion. The GrC-ANN model with the narrowest bandwidth of estimated uncertainty (bandwidth-factor = 0.56) that brackets the highest percentage of true Dx data (i.e., 100%) is the best model to compute Dx in streams. Considering the significant inherent uncertainty reported in the previous Dx models, the GrC-ANN model developed in this study is shown to have a robust performance for evaluating pollutant mixing (Dx) in turbulent environmental flow systems.

Simulation of infiltration facilities using the SEEP/W model and quantification of flood runoff reduction effect by the decrease in CN.

In this study, four infiltration facilities (permeable pavement, infiltration gutter, infiltration trench, and infiltration well) have been investigated and compared with their flood runoff reduction effect. The SEEP/W model was used to estimate the infiltration amount of each facility, and the flood runoff reduction effect was quantified by the decrease in curve number (CN). As a result of this study, we found that: (1) the infiltration could be successfully simulated by the SEEP/W model, whose result could also be quantified effectively by the decrease in CN; (2) among the four infiltration facilities considered in this study, the infiltration well and infiltration trench were found to be most efficient and economical; (3) finally, the intervention effect of the nearby infiltration facility was found not so significant. In an extreme case where the infiltration wells were located at 1 m interval, the intervention effect was found to be just 1%.