Predicting hourly air pollutant levels using artificial neural networks coupled with uncertainty analysis by Monte Carlo simulations

Recent progress in developing artificial neural network (ANN) metamodels has paved the way for reliable use of these models in the prediction of air pollutant concentrations in urban atmosphere. However, improvement of prediction performance, proper selection of input parameters and model architecture, and quantification of model uncertainties remain key challenges to their practical use. This study has three main objectives: to select an ensemble of input parameters for ANN metamodels consisting of meteorological variables that are predictable by conventional weather forecast models and variables that properly describe the complex nature of pollutant source conditions in a major city, to optimize the ANN models to achieve the most accurate hourly prediction for a case study (city of Tehran), and to examine a methodology to analyze uncertainties based on ANN and Monte Carlo simulations (MCS). In the current study, the ANNs were constructed to predict criteria pollutants of nitrogen oxides (NO_x), nitrogen dioxide (NO₂), nitrogen monoxide (NO), ozone (O₃), carbon monoxide (CO), and particulate matter with aerodynamic diameter of less than 10 μm (PM₁₀) in Tehran based on the data collected at a monitoring station in the densely populated central area of the city. The best combination of input variables was comprehensively investigated taking into account the predictability of meteorological input variables and the study of model performance, correlation coefficients, and spectral analysis. Among numerous meteorological variables, wind speed, air temperature, relative humidity and wind direction were chosen as input variables for the ANN models. The complex nature of pollutant source conditions was reflected through the use of hour of the day and month of the year as input variables and the development of different models for each day of the week. After that, ANN models were constructed and validated, and a methodology of computing prediction intervals (PI) and probability of exceeding air quality thresholds was developed by combining ANNs and MCSs based on Latin Hypercube Sampling (LHS). The results showed that proper ANN models can be used as reliable metamodels for the prediction of hourly air pollutants in urban environments. High correlations were obtained with R ² of more than 0.82 between modeled and observed hourly pollutant levels for CO, NO_x, NO₂, NO, and PM₁₀. However, predicted O₃ levels were less accurate. The combined use of ANNs and MCSs seems very promising in analyzing air pollution prediction uncertainties. Replacing deterministic predictions with probabilistic PIs can enhance the reliability of ANN models and provide a means of quantifying prediction uncertainties.     

Variation of surface ozone in Campo Grande,Brazil: meteorological effect analysis and prediction

The effect of meteorological variables on surface ozone (O₃) concentrations was analysed based on temporal variation of linear correlation and artificial neural network (ANN) models defined by genetic algorithms (GAs). ANN models were also used to predict the daily average concentration of this air pollutant in Campo Grande, Brazil. Three methodologies were applied using GAs, two of them considering threshold models. In these models, the variables selected to define different regimes were daily average O₃ concentration, relative humidity and solar radiation. The threshold model that considers two O₃ regimes was the one that correctly describes the effect of important meteorological variables in O₃ behaviour, presenting also a good predictive performance. Solar radiation, relative humidity and rainfall were considered significant for both O₃ regimes; however, wind speed (dispersion effect) was only significant for high concentrations. According to this model, high O₃ concentrations corresponded to high solar radiation, low relative humidity and wind speed. This model showed to be a powerful tool to interpret the O₃ behaviour, being useful to define policy strategies for human health protection regarding air pollution.     

Impacts of Meteorological Factors on PM10: Artificial Neural Networks (ANN) and Multiple Linear Regression (MLR) Approaches

In this study, prediction capacities of multi-linear regression (MLR) and artificial neural networks (ANN) onto coarse particulate matter (PM₁₀) concentrations were investigated. Different meteorological factors on particulate pollution were chosen for operating variables in the model analyses. Two different regions (urban and industrial) were identified in the region of Kocaeli, Turkey. All data sets were obtained from air quality monitoring network of the Ministry of Environment and Urban Planning, and 120 data sets were used in the MLR and ANN models. Regression equations explained the effects of the meteorological factors in MLR analyses. In the ANN model, backpropagation network with two hidden layers has achieved the best prediction efficiency. Determination coefficients and error values were examined for each model. ANN models displayed more accurate results compared to MLR.     

Comparison of hybrid machine learning methods for the prediction of short-term meteorological droughts of Sakarya Meteorological Station in Turkey

Drought is a harmful natural disaster with various negative effects on many aspects of life. In this research, short-term meteorological droughts were predicted with hybrid machine learning models using monthly precipitation data (1960–2020 period) of Sakarya Meteorological Station, located in the northwest of Turkey. Standardized precipitation index (SPI), depending only on precipitation data, was used as the drought index, and 1-, 3-, and 6-month time scales for short-term droughts were considered. In the prediction models, drought index was predicted at t?+?1 output variable by using t, t???1, t???2, and t???3 input variables. Artificial neural networks (ANNs), adaptive neuro-fuzzy inference system (ANFIS), Gaussian process regression (GPR), support vector machine regression (SVMR), k-nearest neighbors (KNN) algorithms were employed as stand-alone machine learning methods. Variation mode decomposition (VMD), discrete wavelet transform (DWT), and empirical mode decomposition (EMD) were utilized as pre-processing techniques to create hybrid models. Six different performance criteria were used to assess model performance. The hybrid models used together with the pre-processing techniques were found to be more successful than the stand-alone models. Hybrid VMD-GPR model yielded the best results (NSE?=?0.9345, OI?=?0.9438, R²?=?0.9367) for 1-month time scale, hybrid VMD-GPR model (NSE?=?0.9528, OI?=?0.9559, R²?=?0.9565) for 3-month time scale, and hybrid DWT-ANN model (NSE?=?0.9398, OI?=?0.9483, R²?=?0.9450) for 6-month time scale. Considering the entire performance criteria, it was determined that the decomposition success of VMD was higher than DWT and EMD.

     

Climatology and forecast modeling of ambient carbon monoxide in Phoenix, Arizona

We perform a climatology of factors influencing ambient carbon monoxide (CO), in which we examine the relationships between meteorology, traffic patterns, and CO at seasonal, weekly, and diurnal time scales in Phoenix, Arizona. From this analysis we identify a range of potentially important variables for statistical CO modeling. Using stepwise multivariate regression, we create a suite of models for hourly and 8-h ambient CO designed for daily operational forecasting purposes. The resulting models include variables and interaction terms related to anticipated nocturnal atmospheric stability as well as antecedent and climatological CO behavior. The models are evaluated using a range of error statistics and skill measures. The most successful approach employs a two-stage modeling strategy in which an initial prediction is made that may, depending on the forecast value, be followed by a second prediction that improves upon the first. The best models provide accurate daily forecasts of CO, with explained variances approaching 0.9 and errors under 1 ppm.     

Ozone levels in the empty quarter of Saudi Arabia—application of adaptive neuro-fuzzy model

In arid regions, primary pollutants may contribute to the increase of ozone levels and cause negative effects on biotic health. This study investigates the use of adaptive neuro-fuzzy inference system (ANFIS) for ozone prediction. The initial fuzzy inference system is developed by using fuzzy C-means (FCM) and subtractive clustering (SC) algorithms, which determines the important rules, increases generalization capability of the fuzzy inference system, reduces computational needs, and ensures speedy model development. The study area is located in the Empty Quarter of Saudi Arabia, which is considered as a source of huge potential for oil and gas field development. The developed clustering algorithm-based ANFIS model used meteorological data and derived meteorological data, along with NO and NO₂ concentrations and their transformations, as inputs. The root mean square error and Willmott’s index of agreement of the FCM- and SC-based ANFIS models are 3.5 ppbv and 0.99, and 8.9 ppbv and 0.95, respectively. Based on the analysis of the performance measures and regression error characteristic curves, it is concluded that the FCM-based ANFIS model outperforms the SC-based ANFIS model.     

Further validation of artificial neural network-based emissions simulation models for conventional and hybrid electric vehicles

With the advent of hybrid electric vehicles, computer-based vehicle simulation becomes more useful to the engineer and designer trying to optimize the complex combination of control strategy, power plant, drive train, vehicle, and driving conditions. With the desire to incorporate emissions as a design criterion, researchers at West Virginia University have developed artificial neural network (ANN) models for predicting emissions from heavy-duty vehicles. The ANN models were trained on engine and exhaust emissions data collected from transient dynamometer tests of heavy-duty diesel engines then used to predict emissions based on engine speed and torque data from simulated operation of a tractor truck and hybrid electric bus. Simulated vehicle operation was performed with the ADVISOR software package. Predicted emissions (carbon dioxide [CO2] and oxides of nitrogen [NO(x)]) were then compared with actual emissions data collected from chassis dynamometer tests of similar vehicles. This paper expands on previous research to include different driving cycles for the hybrid electric bus and varying weights of the conventional truck. Results showed that different hybrid control strategies had a significant effect on engine behavior (and, thus, emissions) and may affect emissions during different driving cycles. The ANN models underpredicted emissions of CO2 and NO(x) in the case of a class-8 truck but were more accurate as the truck weight increased.     

Monthly water quality forecasting and uncertainty assessment via bootstrapped wavelet neural networks under missing data for Harbin, China

In this paper, bootstrapped wavelet neural network (BWNN) was developed for predicting monthly ammonia nitrogen (NH⁴⁺–N) and dissolved oxygen (DO) in Harbin region, northeast of China. The Morlet wavelet basis function (WBF) was employed as a nonlinear activation function of traditional three-layer artificial neural network (ANN) structure. Prediction intervals (PI) were constructed according to the calculated uncertainties from the model structure and data noise. Performance of BWNN model was also compared with four different models: traditional ANN, WNN, bootstrapped ANN, and autoregressive integrated moving average model. The results showed that BWNN could handle the severely fluctuating and non-seasonal time series data of water quality, and it produced better performance than the other four models. The uncertainty from data noise was smaller than that from the model structure for NH⁴⁺–N; conversely, the uncertainty from data noise was larger for DO series. Besides, total uncertainties in the low-flow period were the biggest due to complicated processes during the freeze-up period of the Songhua River. Further, a data missing–refilling scheme was designed, and better performances of BWNNs for structural data missing (SD) were observed than incidental data missing (ID). For both ID and SD, temporal method was satisfactory for filling NH⁴⁺–N series, whereas spatial imputation was fit for DO series. This filling BWNN forecasting method was applied to other areas suffering “real” data missing, and the results demonstrated its efficiency. Thus, the methods introduced here will help managers to obtain informed decisions.     

Artificial neural network (ANN) modeling of adsorption of methylene blue by NaOH-modified rice husk in a fixed-bed column system

In this study, rice husk was modified with NaOH and used as adsorbent for dynamic adsorption of methylene blue (MB) from aqueous solutions. Continuous removal of MB from aqueous solutions was studied in a laboratory scale fixed-bed column packed with NaOH-modified rice husk (NMRH). Effect of different flow rates and bed heights on the column breakthrough performance was investigated. In order to determine the most suitable model for describing the adsorption kinetics of MB in the fixed-bed column system, the bed depth service time (BDST) model as well as the Thomas model was fitted to the experimental data. An artificial neural network (ANN)-based model was also developed for describing the dynamic dye adsorption process. An extensive error analysis was carried out between experimental data and data predicted by the models by using the following error functions: correlation coefficient (R ²), average relative error, sum of the absolute error and Chi-square statistic test (χ ²). Results show that with increasing bed height and decreasing flow rate, the breakthrough time was delayed. All the error functions yielded minimum values for the ANN model than the traditional models (BDST and Thomas), suggesting that the ANN model is the most suitable model to describe the fixed-bed adsorption of MB by NMRH. It is also more rational and reliable to interpret dynamic dye adsorption data through a process of ANN architecture.     

Application of artificial neural network for prediction of Pb(II) adsorption characteristics

The adsorption of Pb(II) onto the surface of microwave-assisted activated carbon was studied through a two-layer feedforward neural network. The activated carbon was developed by microwave activation of Acacia auriculiformis scrap wood char. The prepared adsorbent was characterized by using Brunauer–Emmett–Teller (BET) surface area analyzer, scanning electron microscope (SEM), and X-ray difractometer. In the present study, the input variables for the proposed network were solution pH, contact time, initial adsorbate concentration, adsorbent dose and temperature, whereas the output variable was the percent Pb(II) removal. The network had been trained by using different algorithms and based on the lowest mean squared error (MSE) value and validation error, resilient backpropagation algorithm with 12 neurons in the hidden layer was selected for the present investigation. The tan sigmoid and purelin transfer function were used in the hidden and the output layers of the proposed network, respectively. The model predicted and experimental values of the percent Pb(II) removal were also compared and both the values were found to be in reasonable agreement with each other. The performance of the developed network was further improved by normalizing the experimental data set and it was found that after normalization, the MSE and validation error were reduced significantly. The sensitivity analysis was also performed to determine the most significant input parameter.     

Support vector machine―an alternative to artificial neuron network for water quality forecasting in an agricultural nonpoint source polluted river?

Water quality forecasting in agricultural drainage river basins is difficult because of the complicated nonpoint source (NPS) pollution transport processes and river self-purification processes involved in highly nonlinear problems. Artificial neural network (ANN) and support vector model (SVM) were developed to predict total nitrogen (TN) and total phosphorus (TP) concentrations for any location of the river polluted by agricultural NPS pollution in eastern China. River flow, water temperature, flow travel time, rainfall, dissolved oxygen, and upstream TN or TP concentrations were selected as initial inputs of the two models. Monthly, bimonthly, and trimonthly datasets were selected to train the two models, respectively, and the same monthly dataset which had not been used for training was chosen to test the models in order to compare their generalization performance. Trial and error analysis and genetic algorisms (GA) were employed to optimize the parameters of ANN and SVM models, respectively. The results indicated that the proposed SVM models performed better generalization ability due to avoiding the occurrence of overtraining and optimizing fewer parameters based on structural risk minimization (SRM) principle. Furthermore, both TN and TP SVM models trained by trimonthly datasets achieved greater forecasting accuracy than corresponding ANN models. Thus, SVM models will be a powerful alternative method because it is an efficient and economic tool to accurately predict water quality with low risk. The sensitivity analyses of two models indicated that decreasing upstream input concentrations during the dry season and NPS emission along the reach during average or flood season should be an effective way to improve Changle River water quality. If the necessary water quality and hydrology data and even trimonthly data are available, the SVM methodology developed here can easily be applied to other NPS-polluted rivers.     

Fuzzy system models combined with nonlinear regression for daily ground-level ozone predictions

This study focuses on applying a Takagi–Sugeno fuzzy system and a nonlinear regression (NLR) model for ozone predictions in six Kentucky metropolitan areas. The fuzzy “c-means” clustering technique coupled with an optimal output predefuzzification approach (least square method) was used to train the Takagi–Sugeno fuzzy system. The fuzzy system was tuned by specifying the number of rules and the fuzziness factor. The NLR models were based in part on a previously reported, trajectory-based hybrid NLR model that has been used for years for forecasting ground-level ozone in Louisville, KY. The NLR models were each composed of an interactive nonlinear term and several linear terms. Using a common meteorological parameter set as input variables, the NLR models and the Takagi–Sugeno fuzzy systems model exhibited equivalent forecasting performance on test data from 2004. For all 2004 ozone season forecasts for the six metropolitan areas, the mean absolute error was 8.1 ppb for the NLR model and 8.0 ppb for the Takagi–Sugeno fuzzy model. When a nonlinear term (which was part of the NLR model) was included in the fuzzy model, the combined NLR–fuzzy model had slightly better performance than the original NLR model. For all 2004 metropolitan area forecasts, the mean absolute error of the NLR–fuzzy model forecasts was 7.7 ppb. These small differences may be statistically significant, but for practical purposes the performance of the fuzzy models was equivalent to that of the NLR models.     

Modelling of dissolved oxygen content using artificial neural networks: Danube River, North Serbia, case study

The aims of this study are to create an artificial neural network (ANN) model using non-specific water quality parameters and to examine the accuracy of three different ANN architectures: General Regression Neural Network (GRNN), Backpropagation Neural Network (BPNN) and Recurrent Neural Network (RNN), for prediction of dissolved oxygen (DO) concentration in the Danube River. The neural network model has been developed using measured data collected from the Bezdan monitoring station on the Danube River. The input variables used for the ANN model are water flow, temperature, pH and electrical conductivity. The model was trained and validated using available data from 2004 to 2008 and tested using the data from 2009. The order of performance for the created architectures based on their comparison with the test data is RNN > GRNN > BPNN. The ANN results are compared with multiple linear regression (MLR) model using multiple statistical indicators. The comparison of the RNN model with the MLR model indicates that the RNN model performs much better, since all predictions of the RNN model for the test data were within the error of less than ±10 %. In case of the MLR, only 55 % of predictions were within the error of less than ±10 %. The developed RNN model can be used as a tool for the prediction of DO in river waters.     

Performance of ANFIS versus MLP-NN dissolved oxygen prediction models in water quality monitoring

We discuss the accuracy and performance of the adaptive neuro-fuzzy inference system (ANFIS) in training and prediction of dissolved oxygen (DO) concentrations. The model was used to analyze historical data generated through continuous monitoring of water quality parameters at several stations on the Johor River to predict DO concentrations. Four water quality parameters were selected for ANFIS modeling, including temperature, pH, nitrate (NO₃) concentration, and ammoniacal nitrogen concentration (NH₃-NL). Sensitivity analysis was performed to evaluate the effects of the input parameters. The inputs with the greatest effect were those related to oxygen content (NO₃) or oxygen demand (NH₃-NL). Temperature was the parameter with the least effect, whereas pH provided the lowest contribution to the proposed model. To evaluate the performance of the model, three statistical indices were used: the coefficient of determination (R ²), the mean absolute prediction error, and the correlation coefficient. The performance of the ANFIS model was compared with an artificial neural network model. The ANFIS model was capable of providing greater accuracy, particularly in the case of extreme events.     

A framework for error analysis of a long-range transport model with emphasis on parameter uncertainty

A comprehensive framework for model error analysis is applied to the EMEP-W model of longrange transport of sulfur in Europe. This framework includes a proposed taxonomy of model uncertainties. Parameter uncertainties were investigated by Monte Carlo simulation of two source-receptor combinations. A 20% input parameter uncertainty (expressed as a coefficient of variation = standard deviation/mean) yielded a 15–22% output error of total sulfur deposition. The relationship between output error and input uncertainty was approximately proportional. Covariance between parameters can have an important effect on computed model error, and can either exaggerate or reduce errors compared to the uncorrelated case. Of the model state variables, SO₂ air concentration and wet deposition had the highest error, and total sulfur deposition the lowest. It was also found that it is more important to specify the dispersion of the input parameter frequency distributions than their shape. The results of the model error analysis were applied to routine calculations of deposition in Europe. An error (coefficient of variation) of 20% for transfer coefficients throughout Europe yielded spatial variations in the order of a few tens to a few hundreds of km in computed deposition isolines of 2 and 5 g sulfur m⁻²a⁻¹.     

Two-days ahead prediction of daily maximum concentrations of SO2, O3, PM10, NO2, CO in the urban area of Palermo,Italy

Artificial neural networks are functional alternative techniques in modelling the intricate vehicular exhaust emission dispersion phenomenon. Pollutant predictions are notoriously complex when using either deterministic or stochastic models, which explains why this model was developed using a neural network. Neural networks have the ability to learn about non-linear relationships between the used variables. In this paper a recurrent neural network (Elman model) based forecaster for the prediction of daily maximum concentrations of SO₂, O₃, PM10, NO₂, CO in the city of Palermo is proposed. The effectiveness of the presented forecaster was tested using a time series recorded between 1 January 2003 to 31 December 2004 in eight monitoring stations in urban area of Palermo (Italy). Experimental trials show that the developed and tuned model is appropriate, giving small values of root mean square error (RMSE) , mean absolute error (MAE) and mean square error (MSE). In addition, the related correlation coefficient ranges from 0.72 to 0.97 for each forecasted pollutant, underlying a small difference between the forecasted and the measured values. The above results make the proposed forecaster a powerful tool for pollution management systems.     

Modeling and optimization of reductive degradation of chloramphenicol in aqueous solution by zero-valent bimetallic nanoparticles

Purpose

The present study aims to investigate the individual and combined effects of temperature, pH, zero-valent bimetallic nanoparticles (ZVBMNPs) dose, and chloramphenicol (CP) concentration on the reductive degradation of CP using ZVBMNPs in aqueous medium.

Method

Iron?Csilver ZVBMNPs were synthesized. Batch experimental data were generated using a four-factor statistical experimental design. CP reduction by ZVBMNPs was optimized using the response surface modeling (RSM) and artificial neural network-genetic algorithm (ANN-GA) approaches. The RSM and ANN methodologies were also compared for their predictive and generalization abilities using the same training and validation data set. Reductive by-products of CP were identified using liquid chromatography-mass spectrometry technique.

Results

The optimized process variables (RSM and ANN-GA approaches) yielded CP reduction capacity of 57.37 and 57.10?mg?g^?1, respectively, as compared to the experimental value of 54.0?mg?g^?1 with un-optimized variables. The ANN-GA and RSM methodologies yielded comparable results and helped to achieve a higher reduction (>6%) of CP by the ZVBMNPs as compared to the experimental value. The root mean squared error, relative standard error of prediction and correlation coefficient between the measured and model-predicted values of response variable were 1.34, 3.79, and 0.964 for RSM and 0.03, 0.07, and 0.999 for ANN models for the training and 1.39, 3.47, and 0.996 for RSM and 1.25, 3.11, and 0.990 for ANN models for the validation set.

Conclusion

Predictive and generalization abilities of both the RSM and ANN models were comparable. The synthesized ZVBMNPs may be used for an efficient reductive removal of CP from the water.     

Neural Network Model for Predicting Peak Photochemical Pollutant Levels

ABSTRACT

In this paper, an attempt is made for the 24-hr prediction of photochemical pollutant levels using a neural network model. For this purpose, a model is developed that relates peak pollutant concentrations to meteorological and emission variables and indexes. The analysis is based on measurements of O₃ and NO₂ from the city of Athens. The meteorological variables are selected to cover atmospheric processes that determine the fate of the airborne pollutants while special care is taken to ensure the availability of the required input data from routine observations or forecasts. The comparison between model predictions and actual observations shows a good agreement. In addition, a series of sensitivity tests is performed in order to evaluate the sensitivity of the model to the uncertainty in meteorological variables. Model forecasts are generally rather insensitive to small perturbations in most of the input meteorological data, while they are relatively more sensitive in changes in wind speed and direction.     

Interval estimation of urban ozone level and selection of influential factors by employing automatic relevance determination model

In this work, we focus on simulating the ground-level ozone (O3) time series and its daily maximum concentration in Hong Kong urban air by employing the multilayer perceptron (MLP) model combined with the automatic relevance determination (ARD) method (for simplicity, we name it as MLP-ARD model). Two air quality monitoring sites in Hong Kong, i.e., Tsuen Wan and Tung Chung, are selected for the numerical experiments. The MLP-ARD model based on Bayesian evidence framework can provide reliable interval estimation of real observation as well as offering efficient strategy to avoid over-fitting. The performance comparisons between MLP-ARD model and traditional artificial neural network (ANN) model based on maximum likelihood indicate that MLP-ARD model is more powerful to capture the wild fluctuation of O3 level especially during O3 episodes than the traditional model. Furthermore, it can assess and rank the input variables for the prediction according to their relative importance to the output variable, i.e., the daily maximum O3 concentration in this study. The preliminary experimental results indicate that nitric oxide (NO) and solar radiation are the most important input variables for O3 prediction at both selected sites. In addition, the previous daily maximum O3 level is also important for Tung Chung site. In this regard, MLP-ARD model is a feasible tool to interpret the real physical and chemical process of urban O3 variation.     

Prediction of daily averaged PM10 concentrations by statistical time-varying model

In this study, a time-varying statistical model, TVAREX, was proposed for daily averaged PM₁₀ concentrations forecasting of coastal cities. It is a Kalman filter based autoregressive model with exogenous inputs depending on selected meteorological properties on the day of prediction. The TVAREX model was evaluated and compared to an ANN model, trained with the Levenberg–Marquardt backpropagation algorithm subjected to the same set of inputs. It was found that the error statistics of the TVAREX model in general were comparable to those of the ANN model, but the TVAREX model was more efficient in capturing the PM₁₀ pollution episodes due to its online nature, therefore having an appealing advantage for implementation.