Spatiotemporal assessment of groundwater quality and quantity using geostatistical and ensemble artificial intelligence tools.

The study investigated the spatiotemporal relationship between surface hydrological variables and groundwater quality/quantity using geostatistical and AI tools. AI models were developed to estimate groundwater quality from ground-based measurements and remote sensing images, reducing reliance on laboratory testing. Different Kriging techniques were employed to map ground-based measurements and fill data gaps. The methodology was applied to analyze the Maragheh aquifer in northwest Iran, revealing declining groundwater quality due to industrial. discharges and over-extraction. Spatiotemporal analysis indicated a relationship between groundwater depth/quality, precipitation, and temperature. The Root Mean Square Scaled Error (RMSSE) values for all variables ranged from 0.8508 to 1.1688, indicating acceptable performance of the semivariogram models in predicting the variables. Three AI models, namely Feed-Forward Neural Networks (FFNNs), Support Vector Regression (SVR), and Adaptive Neural Fuzzy Inference System (ANFIS), predicted groundwater quality for wet (June) and dry (October) months using input variables such as groundwater depth, temperature, precipitation, Normalized Difference Vegetation Index (NDVI), and Digital Elevation Model (DEM), with Groundwater Quality Index (GWQI) as the target variable. Ensemble methods were employed to combine the outputs of these models, enhancing performance. Results showed strong predictive capabilities, with coefficient of determination values of 0.88 and 0.84 for wet and dry seasons. Ensemble models improved performance by up to 6% and 12% for wet and dry seasons, respectively, potentially advancing groundwater quality modeling in the future.

A fuzzy logic-based approach for groundwater vulnerability assessment.

Groundwater vulnerability assessment systems have been developed to protect groundwater resources. The DRASTIC model calculates the vulnerability index of the aquifer based on seven effective parameters. The application of expert opinion in rating and weighting parameters is the DRASTIC model's major weakness, which increases uncertainty. This study developed a Mamdani fuzzy logic (MFL) in combination with data mining to handle this uncertainty and predict the specific vulnerability. To highlight this approach, the susceptibility of the Qorveh-Dehgolan plain (QDP) and the Ardabil plain aquifers was investigated. The DRASTIC index was calculated between 63 and 160 for the Ardabil plain and between 39 and 146 for the QDP. Despite some similarities between vulnerability maps and nitrate concentration maps, the results of the DRASTIC model based on nitrate concentration cannot be verified according to Heidke skill score (HSS) and total accuracy (TA) criteria. Then the MFL was developed in two scenarios; the first included all seven parameters, whereas the second used only four parameters of the DRASTIC model. The results showed that, in the first scenario of the MFL modeling, TA and HSS values were respectively 0.75 and 0.51 for the Ardabil plain and 0.45 and 0.33 for the QDP. In addition, according to the TA and HSS values, the proposed model was more reliable and practical in groundwater vulnerability assessment than the traditional method, even using four input data.

Assessing long-term climate change impact on spatiotemporal changes of groundwater level using autoregressive-based and ensemble machine learning models.

To evaluate the long-term climate change impacts on groundwater fluctuations of the Ardabil plain, Iran, a groundwater level (GWL) modeling was proposed in this study. Accordingly, the outputs of Global Climate Models (GCMs) under the sixth report of Coupled Model Intercomparison Project (CMIP6) and future scenario of the Shared Socioeconomic Pathway 5-8.5 (SSP5-8.5), were used as climate change forcing to the Machine learning (ML) models. The GCM data were first downscaled and projected for the future via Artificial Neural Networks (ANNs). Based on the results, compared to 2014 (the last year of the base period), the mean annual temperature may increase by 0.8 °C per decade until 2100. On the other hand, the mean precipitation may decrease by about 8% compared to the base period. Then, the centroid wells of clusters were modeled by Feedforward Neural Network (FFNN), examining different input combination sets to simulate both autoregressive and non-autoregressive models. Since each of the ML models can extract different kinds of information from a dataset, after finding the dominant input set via FFNN, GWL time series were modeled via various ML methods. The modeling results indicated that the ensemble of shallow ML models could lead to a 6% more accurate outcome than the individual shallow ML models, and 4% than the deep learning models. Also, the simulation results for future GWLs illustrated that temperature can impact groundwater oscillations directly, whereas precipitation may not have uniform impacts on the GWLs. The uncertainty evolving in the modeling process was quantified and observed to be in acceptable range. Modeling results showed that the main reason for the declining GWL in the Ardabil plain could be primarily linked to the excessive exploitation of the water table, while climate change impact could be also notable.

Spatio-temporal analysis of climate and irrigated vegetation cover changes and their role in lake water level depletion using a pixel-based approach and canonical correlation analysis.

Lake Urmia, located in northwest Iran, was among the world's largest hypersaline lakes but has now experienced a 7 m decrease in water level, from 1278 m to 1271 over 1996 to 2019. There is doubt as to whether the pixel-based analysis (PBA) approach's answer to the lake's drying is a natural process or a result of human intervention. Here, a non-parametric Mann-Kendall trend test was applied to a 21-year record (2000-2020) of satellite data products, i.e., temperature, precipitation, snow cover, and irrigated vegetation cover (IVC). The Google Earth Engine (GEE) cloud-computing platform utilized over 10 sub-basins in three provinces surrounding Lake Urmia to obtain and calculate pixel-based monthly and seasonal scales for the products. Canonical correlation analysis was employed in order to understand the correlation between variables and lake water level (LWL). The trend analysis results show significant increases in temperature (from 1 to 2 °C during 2000-2020) over May-September, i.e., in 87 %-25 % of the basin. However, precipitation has seen an insignificant decrease (from 3 to 9 mm during 2000-2019) in the rainy months (April and May). Snow cover has also decreased and, when compared with precipitation, shows a change in precipitation patterns from snow to rain. IVC has increased significantly in all sub-basins, especially the southern parts of the lake, with the West province making the largest contribution to the development of IVC. According to the PBA, this analysis underpins the very high contribution of IVC to the drying of the lake in more detail, although the contribution of climate change in this matter is also apparent. The development of IVC leads to increased water consumption through evapotranspiration and excess evaporation caused by the storage of water for irrigation. Due to the decreased runoff caused by consumption exceeding the basin's capacity, the lake cannot be fed sufficiently.

The detection and monitoring of pollution caused by gold mining using a vegetation cover index.

This study explores how a vegetation cover (VC) index can be employed as a pollution warning tool in gold mining areas in the Northwest of Iran. The analysis included the following: (a) the extraction of normalized difference vegetation index (NDVI) maps from Landsat images in three zones, i.e., mining operations, upstream areas without any exploration, and the downstream area of the mining activities, (b) calculation of the zones' VC, (c) investigation of transformation trends in each pixel of VC time series using the Mann-Kendall trend test, (d) determination of the pixels with significant VC reduction and the significant starting points of the trend using the sequential Mann-Kendall test, (e) assessment of the correlation between the zones with significantly reduced VC, and (f) a correlation test between average monthly and annual climate parameters and VC. Our results indicate that although 51 ha of VC has been demolished around the mining activities areas (i.e., zone 1), an overall upward trend in vegetation with no chemical leakage is observed into the downstream area of the basin (i.e., zone 3). This upward trend can be mostly attributed to the increasing precipitation and decreasing temperature in the study period. The fact that the area downstream of the mine shows that the heap leaching method for gold mining in Andaryan mine is currently not damaging the vegetation, this likely means that there is no leakage to the surrounding environment from the mine. Our results further show that using NDVI in a pixel-based scale and statistical methods has a high potential to quantify the effects of human activities on surface biophysical characteristics.

Development of artificial intelligence models for well groundwater quality simulation: Different modeling scenarios.

Groundwater is one of the most important freshwater resources, especially in arid and semi-arid regions where the annual amounts of precipitation are small with frequent drought durations. Information on qualitative parameters of these valuable resources is very crucial as it might affect its applicability from agricultural, drinking, and industrial aspects. Although geo-statistics methods can provide insight about spatial distribution of quality factors, applications of advanced artificial intelligence (AI) models can contribute to produce more accurate results as robust alternative for such a complex geo-science problem. The present research investigates the capacity of several types of AI models for modeling four key water quality variables namely electrical conductivity (EC), sodium adsorption ratio (SAR), total dissolved solid (TDS) and Sulfate (SO4) using dataset obtained from 90 wells in Tabriz Plain, Iran; assessed by k-fold testing. Two different modeling scenarios were established to make simulations using other quality parameters and the geographical information. The obtained results confirmed the capabilities of the AI models for modeling the well groundwater quality variables. Among all the applied AI models, the developed hybrid support vector machine-firefly algorithm (SVM-FFA) model achieved the best predictability performance for both investigated scenarios. The introduced computer aid methodology provided a reliable technology for groundwater monitoring and assessment.

Integration of hard and soft supervised machine learning for flood susceptibility mapping.

Flooding is a destructive natural phenomenon that causes many casualties and property losses in different parts of the world every year. Efficient flood susceptibility mapping (FSM) can reduce the risk of this hazard, and has become the main approach in flood risk management. In this study, we evaluated the prediction ability of artificial neural network (ANN) algorithms for hard and soft supervised machine learning classification in FSM by using three ANN algorithms (multi-layer perceptron (MLP), fuzzy adaptive resonance theory (FART), self-organizing map (SOM)) with different activation functions (sigmoidal (-S), linear (-L), commitment (-C), typicality (-T)). We used integration of these models for predicting the spatial expansion probability of flood events in the Ajichay river basin, northwest Iran. Inputs to the ANN were spatial data on 10 flood influencing factors (elevation, slope, aspect, curvature, stream power index, topographic wetness index, lithology, land use, rainfall, and distance to the river). The FSMs obtained as model outputs were trained and tested using flood inventory datasets earned based on previous records of flood damage in the region for the Ajichay river basin. Sensitivity analysis using one factor-at-a-time (OFAT) and all factors-at-a-time (AFAT) demonstrated that all influencing factors had a positive impact on modeling to generate FSM, with altitude having the greatest impact and curvature the least. Model validation was carried out using total operating characteristic (TOC) with an area under the curve (AUC). The highest success rate was found for MLP-S (92.1%) and the lowest for FART-T (75.8%). The projection rate in the validation of FSMs produced by MLP-S, MLP-L, FART-C, FART-T, SOM-C, and SOM-T was found to be 90.1%, 89.6%, 71.7%, 70.8%, 83.8%, and 81.1%, respectively. While integration of hard and soft supervised machine learning classification with activation functions of MLP-S and MLP-L showed a strong flood prediction capability for proper planning and management of flood hazards, MLP-S is a promising method for predicting the spatial expansion probability of flood events.

A novel multi-model data-driven ensemble approach for the prediction of particulate matter concentration.

Accuracy in the prediction of the particulate matter (PM2.5 and PM10) concentration in the atmosphere is essential for both its monitoring and control. In this study, a novel neuro fuzzy ensemble (NF-E) model was proposed for prediction of hourly PM2.5 and PM10 concentration. The NF-E involves careful selection of relevant input parameters for base modelling and using an adaptive neuro fuzzy inference system (ANFIS) model as a nonlinear kernel for obtaining ensemble output. The four base models used include ANFIS, artificial neural network (ANN), support vector regression (SVR) and multilinear regression (MLR). The dominant input parameters for developing the base models were selected using two nonlinear approaches (mutual information and single-input single-output ANN-based sensitivity analysis) and a conventional Pearson correlation coefficient. The NF-E model was found to predict both PM2.5 and PM10 with higher generalization ability and least error. The NF-E model outperformed all the single base models and other linear ensemble techniques with a Nash-Sutcliffe efficiency (NSE) of 0.9594 and 0.9865, mean absolute error (MAE) of 1.63 µg/m3 and 1.66 µg/m3 and BIAS of 0.0760 and 0.0340 in the testing stage for PM2.5 and PM10, respectively. The NF-E could improve the efficiency of other models by 4-22% for PM2.5 and 3-20% for PM10 depending on the model.

Evaluation of the effluent quality parameters of wastewater treatment plant based on uncertainty analysis and post-processing approaches (case study).

Wastewater treatment plants (WWTPs) are highly complicated and dynamic systems and so their appropriate operation, control, and accurate simulation are essential. The simulation of WWTPs according to the process complexity has become an important issue in growing environmental awareness. In recent decades, artificial intelligence approaches have been used as effective tools in order to investigate environmental engineering issues. In this study, the effluent quality of Tabriz WWTP was assessed using two intelligence models, namely support Vector Machine (SVM) and artificial neural network (ANN). In this regard, several models were developed based on influent variables and tested via SVM and ANN methods. Three time scales, daily, weekly, and monthly, were investigated in the modeling process. On the other hand, since applied methods were sensitive to input variables, the Monte Carlo uncertainty analysis method was used to investigate the best-applied model dependability. It was found that both models had an acceptable degree of uncertainty in modeling the effluent quality of Tabriz WWTP. Next, ensemble approaches were applied to improve the prediction performance of Tabriz WWTP. The obtained results comparison showed that the ensemble methods represented better efficiency than single approaches in predicting the performance of Tabriz WWTP.

An emotional artificial neural network for prediction of vehicular traffic noise.

Road traffic is a leading source of environmental noise pollution in large cities, which greatly affects the health and well-being of people. A reliable method for the prediction of road traffic noise is required for monitoring and assessment of traffic noise exposure. This study presents the first application of the Emotional Artificial Neural Network (EANN), as a new generation of neural network method for modeling the road traffic noise in Nicosia, North Cyprus. The efficiency of the EANN model was validated in comparison with the classical feed-forward neural network (FFNN) using two different scenarios with different input combinations. In the first scenario, vehicular classification (the number of cars, medium vehicles, heavy vehicles) and average speed were considered as the models' inputs. In the second scenario, the total traffic and percentage of heavy vehicles were used instead of the classification where the input parameters were total traffic volume, average speed and percentage of heavy vehicles. Application of the EANN model in the prediction of road traffic noise could improve the efficiency of the FFNN, MLR and empirical models at the verification stage up to 14%, 35% and 37%, respectively. Classifying the traffic volume into sub-classes (in scenario 1) before feeding them into the models improved the performance of the EANN and FFNN models at the verification stage by 8% and 12%, respectively. Sensitivity analysis of the input parameters indicated that total traffic volume is the most relevant factor influencing road traffic noise in the study area followed by the number of cars, medium vehicles, heavy vehicles, average speed and percentage of heavy vehicles, respectively.

Artificial intelligence based ensemble model for prediction of vehicular traffic noise.

Vehicular traffic noise is the main source of noise pollution in major cities around the globe. A reliable and accurate method for the estimation of vehicular traffic noise is therefore essential for creating a healthy noise-free environment. In this study, 2 linear (simple average and weighted average) and 2-nonlinear (neural network and neuro-fuzzy) ensemble models were developed by combining the outputs of three Artificial Intelligence (AI) based non-linear models; Adaptive Neuro Fuzzy Inference System (ANFIS), Feed Forward Neural Network (FFNN), Support Vector Regression (SVR) and one Multilinear regression (MLR) model to enhance the performance of the single black box models in predicting vehicular traffic noise of Nicosia city, North Cyprus. In this way, first a nonlinear sensitivity analysis was applied to select the most relevant and dominant input parameters of the traffic data obtained from 12 observation points in the study area. The most dominant parameters in order of their importance were determined to be number of cars, number of van/pickups, number of trucks, average speed and number of buses. Classifying the number of vehicles into five categories before feeding the traffic data into the AI models was observed to improve performance of the single models up to 29% in the verification phase. Out of the four ensembles models developed, the nonlinear ANFIS ensemble was found to be the most robust by improving the performance of ANFIS, FFNN, SVR and MLR models in the verification stage by 11%, 19%, 21% and 31%, respectively.

Rainfall time series disaggregation in mountainous regions using hybrid wavelet-artificial intelligence methods.

In mountainous regions, rainfall can be extremely variable in space and time. The need to simulate rainfall time series at different scales on one hand and the lack of recording such parameters in small scales because of administrative and economic problems, on the other hand, disaggregation of rainfall time series to the desired scale is an essential topic for hydro-environmental studies of such mountainous regions. Hybrid models development by combining data-driven methods of least square support vector machine (LSSVM) and Artificial Neural Network (ANN) and wavelet decomposition for disaggregation of rainfall time series are the purpose of this paper. In this study, for disaggregating the Tabriz and Sahand rain-gauges time series, according to nonlinear characteristics of observed time scales, wavelet-least square support vector machine (WLSSVM) and wavelet-artificial neural network (WANN) hybrid models were proposed. For this purpose, daily data of four rain-gauges and monthly data of six rain-gauges from mountainous basin of the Urmia Lake for seventeen years were decomposed with wavelet transform and then using mutual information and correlation coefficient criteria, the sub-series were ranked and superior sub-series were used as input data of LSSVM and ANN models for disaggregating the monthly rainfall time series to the daily time series. Results obtained by these hybrid disaggregation models were compared with the results of LSSVM, ANN and classic multiple linear regression (MLR) models. The efficiency of WANN model with regard to the WLSSVM, ANN, LSSVM and MLR models at validation stage in the optimized case for Tabriz rain-gauge showed up to 9.1%, 22%, 20% and 50% increase and in the optimized case for Sahand rain-gauge showed up to 4.5%, 21.1%, 30.2% and 53.3% increase, respectively.

Wastewater treatment plant performance analysis using artificial intelligence - an ensemble approach.

In the present study, three different artificial intelligence based non-linear models, i.e. feed forward neural network (FFNN), adaptive neuro fuzzy inference system (ANFIS), support vector machine (SVM) approaches and a classical multi-linear regression (MLR) method were applied for predicting the performance of Nicosia wastewater treatment plant (NWWTP), in terms of effluent biological oxygen demand (BODeff), chemical oxygen demand (CODeff) and total nitrogen (TNeff). The daily data were used to develop single and ensemble models to improve the prediction ability of the methods. The obtained results of single models proved that, ANFIS model provides effective outcomes in comparison with single models. In the ensemble modeling, simple averaging ensemble, weighted averaging ensemble and neural network ensemble techniques were proposed subsequently to improve the performance of the single models. The results showed that in prediction of BODeff, the ensemble models of simple averaging ensemble (SAE), weighted averaging ensemble (WAE) and neural network ensemble (NNE), increased the performance efficiency of artificial intelligence (AI) modeling up to 14%, 20% and 24% at verification phase, respectively, and less than or equal to 5% for both CODeff and TNeff in calibration phase. This shows that NNE model is more robust and reliable ensemble method for predicting the NWWTP performance due to its non-linear averaging kernel.

Experimental and AI-based numerical modeling of contaminant transport in porous media.

This study developed a new hybrid artificial intelligence (AI)-meshless approach for modeling contaminant transport in porous media. The key innovation of the proposed approach is that both black box and physically-based models are combined for modeling contaminant transport. The effectiveness of the approach was evaluated using experimental and real world data. Artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) were calibrated to predict temporal contaminant concentrations (CCs), and the effect of noisy and de-noised data on the model performance was evaluated. Then, considering the predicted CCs at test points (TPs, in experimental study) and piezometers (in Myandoab plain) as interior conditions, the multiquadric radial basis function (MQ-RBF), as a meshless approach which solves partial differential equation (PDE) of contaminant transport in porous media, was employed to estimate the CC values at any point within the study area where there was no TP or piezometer. Optimal values of the dispersion coefficient in the advection-dispersion PDE and shape coefficient of MQ-RBF were determined using the imperialist competitive algorithm. In temporal contaminant transport modeling, de-noised data enhanced the performance of ANN and ANFIS methods in terms of the determination coefficient, up to 6 and 5%, respectively, in the experimental study and up to 39 and 18%, respectively, in the field study. Results showed that the efficiency of ANFIS-meshless model was more than ANN-meshless model up to 2 and 13% in the experimental and field studies, respectively.

Daily suspended sediment concentration simulation using ANN and neuro-fuzzy models.

In the present study, artificial neural networks (ANNs), neuro-fuzzy (NF), multi linear regression (MLR) and conventional sediment rating curve (SRC) models are considered for time series modeling of suspended sediment concentration (SSC) in rivers. As for the artificial intelligence systems, feed forward back propagation (FFBP) method and Sugeno inference system are used for ANNs and NF models, respectively. The models are trained using daily river discharge and SSC data belonging to Little Black River and Salt River gauging stations in the USA. Obtained results demonstrate that ANN and NF models are in good agreement with the observed SSC values; while they depict better results than MLR and SRC methods. For example, in Little Black River station, the determination coefficient is 0.697 for NF model, while it is 0.457, 0.257 and 0.225 for ANN, MLR and SRC models, respectively. The values of cumulative suspended sediment load estimated by ANN and NF models are closer to the observed data than the other models. In general, the results illustrate that NF model presents better performance in SSC prediction in compression to other models.