Scrutinization of land subsidence rate using a supportive predictive model: Incorporating radar interferometry and ensemble soft-computing.

Land subsidence is a huge challenge that land and water resource managers are still facing. Radar datasets revolutionize the way and give us the ability to provide information about it, thanks to their low cost. But identifying the most important drivers need for the modeling process. Machine learning methods are especially top of mind amid the prediction studies of natural hazards and hit new heights over the last couple of years. Hence, putting an efficient approach like integrated radar-and-ensemble-based method into practice for land subsidence rate simulation is not available yet which is the main aim of this research. In this study, the number of 52 pairs of radar images were used to identify subsidence from 2014 to 2019. Then, using the simulated annealing (SA) algorithm the key variables affecting land subsidence were identified among the topographical parameters, aquifer information, land use, hydroclimatic variables, and geological and soil factors. Afterward, three individual machine learning models (including Support Vector Machine, SVM; Gaussian Process, GP; Bayesian Additive Regression Tree, BART) along with three ensemble learning approaches were considered for land subsidence rate modeling. The results indicated that the subsidence varies between 0 and 59 cm in this period. Comparing the Radar results with the permanent geodynamic station exhibited a very strong correlation between the ground station and the radar images (R2 = 0.99, RMSE = 0.008). Parsing the input data by the SA indicated that key drivers are precipitation, elevation, percentage of fine-grained materials in the saturated zone, groundwater withdrawal, distance to road, groundwater decline, and aquifer thickness. The performance comparison indicated that ensemble models perform better than individual models, and among ensemble models, the nonlinear ensemble approach (i.e., BART model combination) provided better performance (RMSE = 0.061, RSR = 0.42, R2 = 0.83, PBIAS = 2.2). Also, the distribution shape of the probability density function in the non-linear ensemble model is much closer to the observations. Results indicated that the presence of significant fine-grained materials in unconsolidated aquifer systems can clarify the response of the aquifer system to groundwater decline, low recharge, and subsequent land subsidence. Therefore, the interaction between these factors can be very dangerous and intensify subsidence.

Newly explored machine learning model for river flow time series forecasting at Mary River, Australia.

Hourly river flow pattern monitoring and simulation is the indispensable precautionary task for river engineering sustainability, water resource management, flood risk mitigation, and impact reduction. Reliable river flow forecasting is highly emphasized to support major decision-makers. This research paper adopts a new implementation approach for the application of a river flow prediction model for hourly prediction of the flow of Mary River in Australia; a novel data-intelligent model called emotional neural network (ENN) was used for this purpose. A historical dataset measured over a 4-year period (2011-2014) at hourly timescale was used in building the ENN-based predictive model. The results of the ENN model were validated against the existing approaches such as the minimax probability machine regression (MPMR), relevance vector machine (RVM), and multivariate adaptive regression splines (MARS) models. The developed models are evaluated against each other for validation purposes. Various numerical and graphical performance evaluators are conducted to assess the predictability of the proposed ENN and the competitive benchmark models. The ENN model, used as an objective simulation tool, revealed an outstanding performance when applied for hourly river flow prediction in comparison with the other benchmark models. However, the order of the model, performance wise, is ENN > MARS > RVM > MPMR. In general, the present results of the proposed ENN model reveal a promising modeling strategy for the hourly simulation of river flow, and such a model can be explored further for its ability to contribute to the state-of-the-art of river engineering and water resources monitoring and future prediction at near real-time forecast horizons.

Earth fissure hazard prediction using machine learning models.

Earth fissures are the cracks on the surface of the earth mainly formed in the arid and the semi-arid basins. The excessive withdrawal of groundwater, as well as the other underground natural resources, has been introduced as the significant causing of land subsidence and potentially, the earth fissuring. Fissuring is rapidly turning into the nations' major disasters which are responsible for significant economic, social, and environmental damages with devastating consequences. Modeling the earth fissure hazard is particularly important for identifying the vulnerable groundwater areas for the informed water management, and effectively enforce the groundwater recharge policies toward the sustainable conservation plans to preserve existing groundwater resources. Modeling the formation of earth fissures and ultimately prediction of the hazardous areas has been greatly challenged due to the complexity, and the multidisciplinary involved to predict the earth fissures. This paper aims at proposing novel machine learning models for prediction of earth fissuring hazards. The Simulated annealing feature selection (SAFS) method was applied to identify key features, and the generalized linear model (GLM), multivariate adaptive regression splines (MARS), classification and regression tree (CART), random forest (RF), and support vector machine (SVM) have been used for the first time to build the prediction models. Results indicated that all the models had good accuracy (>86%) and precision (>81%) in the prediction of the earth fissure hazard. The GLM model (as a linear model) had the lowest performance, while the RF model was the best model in the modeling process. Sensitivity analysis indicated that the hazardous class in the study area was mainly related to low elevations with characteristics of high groundwater withdrawal, drop in groundwater level, high well density, high road density, low precipitation, and Quaternary sediments distribution.

Evaluation of watershed health using Fuzzy-ANP approach considering geo-environmental and topo-hydrological criteria.

Assessment of watershed health and prioritization of sub-watersheds are needed to allocate natural resources and efficiently manage watersheds. Characterization of health and spatial prioritization of sub-watersheds in data scarce regions helps better comprehend real watershed conditions and design and implement management strategies. Previous studies on the assessment of health and prioritization of sub-watersheds in ungauged regions have not considered environmental factors and their inter-relationship. In this regard, fuzzy logic theory can be employed to improve the assessment of watershed health. The present study considered a combination of climate vulnerability (Climate Water Balance), relative erosion rate of surficial rocks, slope weighted K-factor, topographic indices, thirteen morphometric characteristics (linear, areal, and relief aspects), and potential non-point source pollution to assess watershed health, using a new framework which considers the complex linkage between human activities and natural resources. The new framework, focusing on watershed health score (WHS), was employed for the spatial prioritization of 31 sub-watersheds in the Khoy watershed, West Azerbaijan Province, Iran. In this framework, an analytical network process (ANP) and fuzzy theory were used to investigate the inter-relationships between the above mentioned geo-environmental factors and to classify and rank the health of each sub-watershed in four classes. Results demonstrated that only one sub-watershed (C15) fell into the class that was defined as 'a potentially critical zone'. This article provides a new framework and practical recommendations for watershed management agencies with a high level of assurance when there is a lack of reliable hydrometric gauge data.

Mapping of salty aeolian dust-source potential areas: Ensemble model or benchmark models?

Considering the effects of dust on human health, environment, agriculture, and transportation, it is necessary to investigate dust emissions susceptibility. This study aimed to study the capability of different machine learning models in analyzing land susceptibility to dust emissions. At first, the dust-source areas were identified by examining the frequency of occurrence (FOO) of dusty days using the aerosol optical depth (AOD) of the MODIS sensor from 2000 to 2020 and field surveys. Then, the weighted subspace random forest (WSRF) model in comparison with three benchmark models-general linear model (GLM), boosted regression tree (BRT), and support vector machine (SVM)-was employed to predict land susceptibility to dust emissions and also to determine the importance of dust-drivers. The results revealed that the WSRF outperformed benchmark models. In a nutshell, the values of accuracy, Kappa, and probability of detection for all models were more than 97 %, and also the false alarm rate was less than 1 % for all models. Spatial analysis indicated a greater frequency of dust events in the outskirts of Urmia Lake (mainly in the eastern and southern parts). Furthermore, according to the map of land susceptibility to dust emissions produced by the WSRF model, about 4.5 %, 2.8 %, 1.8 %, 0.8 %, and 0.2 % of the salt land, rangeland, agricultural, dry-farming, and barren lands, respectively, associated with high and very high degrees of dust emissions susceptibility. Therefore, this study provided in-depth insights into the applicability of the ensemble model, WSRF, to precisely map dust emissions susceptibility.

Susceptibility Assessment of Groundwater Nitrate Contamination Using an Ensemble Machine Learning Approach.

Groundwater pollution susceptibility mapping using parsimonious approaches with limited data is of utmost importance for water resource and health planning, especially in data-scarce regions. Current research assesses groundwater nitrate susceptibility by considering the various combination of explanatory variables. In this study, the novel machine learning models of weighted subspace random forest (WSRF) and generalized additive model using LOESS (GAMLOESS) are applied, and the results are compared with well-known machine learning models of K-nearest neighbors (KKNN) and random forest (RF). The optimum combination of inputs for groundwater nitrate susceptibility mapping is identified using the k-fold cross-validation methodology. Results indicated that the combination of variables of precipitation, groundwater level, and lithology had the best performance among the 16 combinations. Modeling performance using the optimum combination demonstrated that the new ensemble approach, the WSRF model, had superior performance according to the evaluation metrics of accuracy (0.87), kappa (0.73), precision (0.92), false alarm ratio (0.08), and critical success index (0.75). The susceptibility assessment results of this paper can be a useful tool in developing strategies for the prevention and protection of groundwater pollution.

Towards a flood vulnerability assessment of watershed using integration of decision-making trial and evaluation laboratory, analytical network process, and fuzzy theories.

Among natural disasters, flood is increasingly recognized as a serious worldwide concern that causes the most damages in parts of agriculture, fishery, housing, and infrastructure and strongly affects economic and social activities. Universally, there is a requirement to increase our conception of flood vulnerability and to outstretch methods and tools to assess it. Spatial analysis of flood vulnerability is part of non-structural measures to prevent and reduce flood destructive effects. Hence, the current study proposes a methodology for assessing the flood vulnerability in the area of watershed in a severely flooded area of Iran (i.e., Kashkan Watershed). First interdependency analysis among criteria (including population density (PD), livestock density (LD), percentage of farmers and ranchers (PFR), distance to industrial and mining areas (DTIM), distance to tourist and cultural heritage areas (DTTCH), land use, distance to residential areas (DTRe), distance to road (DTR), and distance to stream (DTS)) was conducted using the decision-making trial and evaluation laboratory (DEMATEL) method. Hence, the cause and effect factors and their interaction levels in the whole network were investigated. Then, using the interdependency relationships among criteria, a network structure from flood vulnerability factors to determine their importance of factors was constructed, and the analytical network process (ANP) was applied. Finally, with the aim to overcome ambiguity, reduce uncertainty, and keep the data variability, an appropriate fuzzy membership function was applied to each layer by analyzing the relationship of each layer with flood vulnerability. Importance analysis indicated that land use (0.197), DTS (0.181), PD (0.180), DTRe (0.140), and DTR (0.138) were the most important variables. The flood vulnerability map produced by the integrated method of DEMATEL-ANP-fuzzy showed that about 19.2% of the region has a high to very high flood vulnerability.

Fuzzy clustering and distributed model for streamflow estimation in ungauged watersheds.

This paper proposes a regionalization method for streamflow prediction in ungauged watersheds in the 7461 km2 area above the Gharehsoo Hydrometry Station in the Ardabil Province, in the north of Iran. First, the Fuzzy c-means clustering method (FCM) was used to divide 46 gauged (19) and ungauged (27) watersheds into homogenous groups based on a variety of topographical and climatic factors. After identifying the homogenous watersheds, the Soil and Water Assessment Tool (SWAT) was calibrated and validated using data from the gauged watersheds in each group. The calibrated parameters were then tested in another gauged watershed that we considered as a pseudo ungauged watershed in each group. Values of R-Squared and Nash-Sutcliffe efficiency (NSE) were both ≥ 0.70 during the calibration and validation phases; and ≥ 0.80 and ≥ 0.74, respectively, during the testing in the pseudo ungauged watersheds. Based on these metrics, the validated regional models demonstrated a satisfactory result for predicting streamflow in the ungauged watersheds within each group. These models are important for managing stream quantity and quality in the intensive agriculture study area.

Accuracy assessment of remotely sensed data to analyze lake water balance in semi-arid region.

Lake water level fluctuation is a function of hydro-meteorological components, namely input, and output to the system. The combination of these components from in-situ and remote sensing sources has been used in this study to define multiple scenarios, which are the major explanatory pathways to assess lake water levels. The goal is to analyze each scenario through the application of the water balance equation to simulate lake water levels. The largest lake in Iran, Lake Urmia, has been selected in this study as it needs a great deal of attention in terms of water management issues. We ran a monthly water balance simulation of nineteen scenarios for Lake Urmia from 2003 to 2007 by applying different combinations of data, including observed and remotely sensed water level, flow, evaporation, and rainfall. We used readily available water level data from Hydrosat, Hydroweb, and DAHITI platforms; evapotranspiration from MODIS and rainfall from TRMM. The analysis suggests that the consideration of field data in the algorithm as the initial water level can reproduce the fluctuation of Lake Urmia water level in the best way. The scenario that combines in-situ meteorological components is the closest match to the observed water level of Lake Urmia. Almost all scenarios showed good dynamics with the field water level, but we found that nine out of nineteen scenarios did not vary significantly in terms of dynamics. The results also reveal that, even without any field data, the proposed scenario, which consists entirely of remote sensing components, is capable of estimating water level fluctuation in a lake. The analysis also explains the necessity of using proper data sources to act on water regulations and managerial decisions to understand the temporal phenomenon not only for Lake Urmia but also for other lakes in semi-arid regions.

Susceptibility mapping of groundwater salinity using machine learning models.

Increasing groundwater salinity has recently raised severe environmental and health concerns around the world. Advancement of the novel methods for spatial salinity modeling and prediction would be essential for effective management of the resources and planning mitigation policies. The current research presents the application of machine learning (ML) models in groundwater salinity mapping based on the dichotomous predictions. The groundwater salinity is predicted using the essential factors (i.e., identified by the simulated annealing feature selection methodology) through k-fold cross-validation methodology. Six ML models, namely, flexible discriminant analysis (FDA), mixture discriminant analysis (MAD), boosted regression tree (BRT), multivariate adaptive regression spline (MARS), random forest (RF), support vector machine (SVM), were employed to groundwater salinity mapping. The results of the modeling indicated that the SVM model had superior performance than other models. Variables of soil order, groundwater withdrawal, precipitation, land use, and elevation had the most contribute to groundwater salinity mapping. Results highlighted that the southern parts of the region and some parts in the north, northeast, and west have a high groundwater salinity, in which these areas are mostly matched with soil order of Entisols, bareland areas, and low elevations.

Spatial hazard assessment of the PM10 using machine learning models in Barcelona, Spain.

Air pollution, and especially atmospheric particulate matter (PM), has a profound impact on human mortality and morbidity, environment, and ecological system. Accordingly, it is very relevant predicting air quality. Although the application of the machine learning (ML) models for predicting air quality parameters, such as PM concentrations, has been evaluated in previous studies, those on the spatial hazard modeling of them are very limited. Due to the high potential of the ML models, the spatial modeling of PM can help managers to identify the pollution hotspots. Accordingly, this study aims at developing new ML models, such as Random Forest (RF), Bagged Classification and Regression Trees (Bagged CART), and Mixture Discriminate Analysis (MDA) for the hazard prediction of PM10 (particles with a diameter less than 10 µm) in the Barcelona Province, Spain. According to the annual PM10 concentration in 75 stations, the healthy and unhealthy locations are determined, and a ratio 70/30 (53/22 stations) is applied for calibrating and validating the ML models to predict the most hazardous areas for PM10. In order to identify the influential variables of PM modeling, the simulated annealing (SA) feature selection method is used. Seven features, among the thirteen features, are selected as critical features. According to the results, all the three-machine learning (ML) models achieve an excellent performance (Accuracy > 87% and precision > 86%). However, the Bagged CART and RF models have the same performance and higher than the MDA model. Spatial hazard maps predicted by the three models indicate that the high hazardous areas are located in the middle of the Barcelona Province more than in the Barcelona's Metropolitan Area.

Mass wasting susceptibility assessment of snow avalanches using machine learning models.

Snow avalanche is among the most harmful natural hazards with major socioeconomic and environmental destruction in the cold and mountainous regions. The devastating propagation and accumulation of the snow avalanche debris and mass wasting of surface rocks and vegetation particles threaten human life, transportation networks, built environments, ecosystems, and water resources. Susceptibility assessment of snow avalanche hazardous areas is of utmost importance for mitigation and development of land-use policies. This research evaluates the performance of the well-known machine learning methods, i.e., generalized additive model (GAM), multivariate adaptive regression spline (MARS), boosted regression trees (BRT), and support vector machine (SVM), in modeling the mass wasting hazard induced by snow avalanches. The key features are identified by the recursive feature elimination (RFE) method and used for the model calibration. The results indicated a good performance of the modeling process (Accuracy > 0.88, Kappa > 0.76, Precision > 0.84, Recall > 0.86, and AUC > 0.89), which the SVM model highlighted superior performance than others. Sensitivity analysis demonstrated that the topographic position index (TPI) and distance to stream (DTS) were the most important variables which had more contribution in producing the susceptibility maps.

Flash-flood hazard assessment using ensembles and Bayesian-based machine learning models: Application of the simulated annealing feature selection method.

Flash-floods are increasingly recognized as a frequent natural hazard worldwide. Iran has been among the mostdevastated regions affected by the major floods. While the temporal flash-flood forecasting models are mainly developed for warning systems, the models for assessing hazardous areas can greatly contribute to adaptation and mitigation policy-making and disaster risk reduction. Former researches in the flash-flood hazard mapping have heightened the urge for the advancement of more accurate models. Thus, the current research proposes the state-of-the-art ensemble models of boosted generalized linear model (GLMBoost) and random forest (RF), and Bayesian generalized linear model (BayesGLM) methods for higher performance modeling. Furthermore, a pre-processing method, namely simulated annealing (SA), is used to eliminate redundant variables from the modeling process. Results of the modeling based on the hit and miss analysis indicates high performance for both models (accuracy = 90-92%, Kappa = 79-84%, Success ratio = 94-96%, Threat score = 80-84%, and Heidke skill score = 79-84%). The variables of distance from the stream, vegetation, drainage density, land use, and elevation have shown more contribution among others for modeling the flash-flood. The results of this study can significantly facilitate mapping the hazardous areas and further assist watershed managers to control and remediate induced damages of flood in the data-scarce regions.

Integrated machine learning methods with resampling algorithms for flood susceptibility prediction.

Flood susceptibility projections relying on standalone models, with one-time train-test data splitting for model calibration, yields biased results. This study proposed novel integrative flood susceptibility prediction models based on multi-time resampling approaches, random subsampling (RS) and bootstrapping (BT) algorithms, integrated with machine learning models: generalized additive model (GAM), boosted regression tree (BTR) and multivariate adaptive regression splines (MARS). RS and BT algorithms provided 10 runs of data resampling for learning and validation of the models. Then the mean of 10 runs of predictions is used to produce the flood susceptibility maps (FSM). This methodology was applied to Ardabil Province on coastal margins of the Caspian Sea which faced destructive floods. The area under curve (AUC) of receiver operating characteristic (ROC) and true skill statistic (TSS) and correlation coefficient (COR) were utilized to evaluate the predictive accuracy of the proposed models. Results demonstrated that resampling algorithms improved the performance of Standalone GAM, MARS and BRT models. Results also revealed that Standalone models had better performance with the BT algorithm compared to the RS algorithm. BT-GAM model attained superior performance in terms of statistical measures (AUC = 0.98, TSS = 0.93, COR = 0.91), followed by BT-MARS (AUC = 0.97, TSS = 0.91, COR = 0.91) and BT-BRT model (AUC = 0.95, TSS = 0.79, COR = 0.79). Results demonstrated that the proposed models outperformed the benchmark models such as Standalone GAM, MARS, BRT, multilayer perceptron (MLP) and support vector machine (SVM). Given the admirable performance of the proposed models in a large scale area, the promising results can be expected from these models for other regions.

An ensemble prediction of flood susceptibility using multivariate discriminant analysis, classification and regression trees, and support vector machines.

Floods, as a catastrophic phenomenon, have a profound impact on ecosystems and human life. Modeling flood susceptibility in watersheds and reducing the damages caused by flooding is an important component of environmental and water management. The current study employs two new algorithms for the first time in flood susceptibility analysis, namely multivariate discriminant analysis (MDA), and classification and regression trees (CART), incorporated with a widely used algorithm, the support vector machine (SVM), to create a flood susceptibility map using an ensemble modeling approach. A flood susceptibility map was developed using these models along with a flood inventory map and flood conditioning factors (including altitude, slope, aspect, curvature, distance from river, topographic wetness index, drainage density, soil depth, soil hydrological groups, land use, and lithology). The case study area was the Khiyav-Chai watershed in Iran. To ensure a more accurate ensemble model, this study proposed a framework for flood susceptibility assessment where only those models with an accuracy of >80% were permissible for use in ensemble modeling. The relative importance of factors was determined using the Jackknife test. Results indicated that the MDA model had the highest predictive accuracy (89%), followed by the SVM (88%) and CART (0.83%) models. Sensitivity analysis showed that slope percent, drainage density, and distance from river were the most important factors in flood susceptibility mapping. The ensemble modeling approach indicated that residential areas at the outlet of the watershed were very susceptible to flooding, and that these areas should, therefore, be prioritized for the prevention and remediation of floods.

A novel bias correction framework of TMPA 3B42 daily precipitation data using similarity matrix/homogeneous conditions.

Reduction of bias in remotely sensed precipitation products is a major challenge in environment modeling, hydrology, and managing the water resources. Various bias correction techniques are applied to reduce the bias from pixel to gauge data. However, a successful methodology to improve bias correction on the daily scale is often challenging and limited. We present a methodology that can be used to correct the daily bias in remote sensing rainfall data, and to demonstrate the Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis (TMPA) 3B42 data was used. The proposed bias correction method is based on the concept of similarity (homogeneous) conditions developed based on the periodicity and different percentile-based precipitation amount, and by identifying the best donor pixel to transfer bias correction factor to a specific ungauged pixel (the receptor pixel) based on the similarity (elevation, latitude, and longitude). Bias correction factors were obtained using the mean bias-removal (MBR) and multiplicative ratio (MR) techniques in the cells of the similarity matrix. The proposed methodology demonstrates a significant removal of bias associated with TMPA 3B42 data sets and it is capable of removing the bias in daily precipitation data on an average by 57% (51%) in the gauged pixels, and 25% (22%) in the ungauged pixels for MBR (MR) method.

Predicting uncertainty of machine learning models for modelling nitrate pollution of groundwater using quantile regression and UNEEC methods.

Although estimating the uncertainty of models used for modelling nitrate contamination of groundwater is essential in groundwater management, it has been generally ignored. This issue motivates this research to explore the predictive uncertainty of machine-learning (ML) models in this field of study using two different residuals uncertainty methods: quantile regression (QR) and uncertainty estimation based on local errors and clustering (UNEEC). Prediction-interval coverage probability (PICP), the most important of the statistical measures of uncertainty, was used to evaluate uncertainty. Additionally, three state-of-the-art ML models including support vector machine (SVM), random forest (RF), and k-nearest neighbor (kNN) were selected to spatially model groundwater nitrate concentrations. The models were calibrated with nitrate concentrations from 80 wells (70% of the data) and then validated with nitrate concentrations from 34 wells (30% of the data). Both uncertainty and predictive performance criteria should be considered when comparing and selecting the best model. Results highlight that the kNN model is the best model because not only did it have the lowest uncertainty based on the PICP statistic in both the QR (0.94) and the UNEEC (in all clusters, 0.85-0.91) methods, but it also had predictive performance statistics (RMSE = 10.63, R2 = 0.71) that were relatively similar to RF (RMSE = 10.41, R2 = 0.72) and higher than SVM (RMSE = 13.28, R2 = 0.58). Determining the uncertainty of ML models used for spatially modelling groundwater-nitrate pollution enables managers to achieve better risk-based decision making and consequently increases the reliability and credibility of groundwater-nitrate predictions.

River suspended sediment modelling using the CART model: A comparative study of machine learning techniques.

Suspended sediment load (SSL) modelling is an important issue in integrated environmental and water resources management, as sediment affects water quality and aquatic habitats. Although classification and regression tree (CART) algorithms have been applied successfully to ecological and geomorphological modelling, their applicability to SSL estimation in rivers has not yet been investigated. In this study, we evaluated use of a CART model to estimate SSL based on hydro-meteorological data. We also compared the accuracy of the CART model with that of the four most commonly used models for time series modelling of SSL, i.e. adaptive neuro-fuzzy inference system (ANFIS), multi-layer perceptron (MLP) neural network and two kernels of support vector machines (RBF-SVM and P-SVM). The models were calibrated using river discharge, stage, rainfall and monthly SSL data for the Kareh-Sang River gauging station in the Haraz watershed in northern Iran, where sediment transport is a considerable issue. In addition, different combinations of input data with various time lags were explored to estimate SSL. The best input combination was identified through trial and error, percent bias (PBIAS), Taylor diagrams and violin plots for each model. For evaluating the capability of the models, different statistics such as Nash-Sutcliffe efficiency (NSE), Kling-Gupta efficiency (KGE) and percent bias (PBIAS) were used. The results showed that the CART model performed best in predicting SSL (NSE=0.77, KGE=0.8, PBIAS<±15), followed by RBF-SVM (NSE=0.68, KGE=0.72, PBIAS<±15). Thus the CART model can be a helpful tool in basins where hydro-meteorological data are readily available.

A novel machine learning-based approach for the risk assessment of nitrate groundwater contamination.

This study aimed to develop a novel framework for risk assessment of nitrate groundwater contamination by integrating chemical and statistical analysis for an arid region. A standard method was applied for assessing the vulnerability of groundwater to nitrate pollution in Lenjanat plain, Iran. Nitrate concentration were collected from 102 wells of the plain and used to provide pollution occurrence and probability maps. Three machine learning models including boosted regression trees (BRT), multivariate discriminant analysis (MDA), and support vector machine (SVM) were used for the probability of groundwater pollution occurrence. Afterwards, an ensemble modeling approach was applied for production of the groundwater pollution occurrence probability map. Validation of the models was carried out using area under the receiver operating characteristic curve method (AUC); values above 80% were selected to contribute in ensembling process. Results indicated that accuracy for the three models ranged from 0.81 to 0.87, therefore all models were considered for ensemble modeling process. The resultant groundwater pollution risk (produced by vulnerability, pollution, and probability maps) indicated that the central regions of the plain have high and very high risk of nitrate pollution further confirmed by the exiting landuse map. The findings may provide very helpful information in decision making for groundwater pollution risk management especially in semi-arid regions.