Uncertainty analysis of developed ANN and ANFIS models in prediction of carbon monoxide daily concentration

This study aims to predict daily carbon monoxide (CO) concentration in the atmosphere of Tehran by means of developed artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) models. Forward selection (FS) and Gamma test (GT) methods are used for selecting input variables and developing hybrid models with ANN and ANFIS. From 12 input candidates, 7 and 9 variables are selected using FS and GT, respectively. Evaluation of developed hybrid models and its comparison with ANN and ANFIS models fed with all input variables shows that both FS and GT techniques reduce not only the output error, but also computational cost due to less inputs. FS–ANN and FS–ANFIS models are selected as the best models considering R², mean absolute error and also developed discrepancy ratio statistics. It is also shown that these two models are superior in predicting pollution episodes. Finally, uncertainty analysis based on Monte-Carlo simulation is carried out for FS–ANN and FS–ANFIS models which shows that FS–ANN model has less uncertainty; i.e. it is the best model which forecasts satisfactorily the trends in daily CO concentration levels.     

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.     

Uncertainty in Nonpoint Source Pollution Models and Associated Risks

The usefulness of water quality simulation models for environmental management is explored with a focus on prediction uncertainty. The specific objective is to demonstrate how the usability of a flow and transport model (here: MACRO) can be enhanced by developing and analyzing its output probability distributions based on input variability. This infiltration-based model was designed to investigate preferential flow effects on pollutant transport. A statistical sensitivity analysis is used to identify the most uncertain input parameters based on model outputs. Probability distribution functions of input variables were determined based on field-measured data obtained under alternative tillage treatments. Uncertainty of model outputs is investigated using a Latin hypercube sampling scheme (LHS) with restricted pairing for model input sampling. Probability density functions (pdfs) are constructed for water flow rate, atrazine leaching rate, total accumulated leaching, and atrazine concentration in percolation water. Results indicate that consideration of input parameter uncertainty produces a 20% higher mean flow rate along with two to three times larger atrazine leaching rate, accumulated leachate, and concentration than that obtained using mean input parameters. Uncertainty in predicted flow rate is small but that in solute transport is an order of magnitude larger than that of corresponding input parameters. Macropore flow is observed to contribute to the variability of atrazine transport results. Overall, the analysis provides a quantification of prediction uncertainty that is found to enhance a user's ability to assess risk levels associated with model predictions.     

Uncertainty in Nonpoint Source Pollution Models and Associated Risks

The usefulness of water quality simulation models for environmental management is explored with a focus on prediction uncertainty. The specific objective is to demonstrate how the usability of a flow and transport model (here: MACRO) can be enhanced by developing and analyzing its output probability distributions based on input variability. This infiltration-based model was designed to investigate preferential flow effects on pollutant transport. A statistical sensitivity analysis is used to identify the most uncertain input parameters based on model outputs. Probability distribution functions of input variables were determined based on field-measured data obtained under alternative tillage treatments. Uncertainty of model outputs is investigated using a Latin hypercube sampling scheme (LHS) with restricted pairing for model input sampling. Probability density functions (pdfs) are constructed for water flow rate, atrazine leaching rate, total accumulated leaching, and atrazine concentration in percolation water. Results indicate that consideration of input parameter uncertainty produces a 20% higher mean flow rate along with two to three times larger atrazine leaching rate, accumulated leachate, and concentration than that obtained using mean input parameters. Uncertainty in predicted flow rate is small but that in solute transport is an order of magnitude larger than that of corresponding input parameters. Macropore flow is observed to contribute to the variability of atrazine transport results. Overall, the analysis provides a quantification of prediction uncertainty that is found to enhance a user's ability to assess risk levels associated with model predictions.     

Uncertainties in predicted ozone concentrations due to input uncertainties for the UAM-V photochemical grid model applied to the July 1995 OTAG domain

The photochemical grid model, UAM-V, has been used by regulatory agencies to make decisions concerning emissions controls, based on studies of the July 1995 ozone episode in the eastern US. The current research concerns the effect of the uncertainties in UAM-V input variables (emissions, initial and boundary conditions, meteorological variables, and chemical reactions) on the uncertainties in UAM-V ozone predictions. Uncertainties of 128 input variables have been estimated and most range from about 20% to a factor of two. 100 Monte Carlo runs, each with new resampled values of each of the 128 input variables, have been made for given sets of median emissions assumptions. Emphasis is on the maximum hourly-averaged ozone concentration during the 12–14 July 1995 period. The distribution function of the 100 Monte Carlo predicted domain-wide maximum ozone concentrations is consistently close to log-normal with a 95% uncertainty range extending over plus and minus a factor of about 1.6 from the median. Uncertainties in ozone predictions are found to be most strongly correlated with uncertainties in the NO₂ photolysis rate. Also important are wind speed and direction, relative humidity, cloud cover, and biogenic VOC emissions. Differences in median predicted maximum ozone concentrations for three alternate emissions control assumptions were investigated, with the result that (1) the suggested year-2007 emissions changes would likely be effective in reducing concentrations from those for the year-1995 actual emissions, that (2) an additional 50% NO_x emissions reductions would likely be effective in further reducing concentrations, and that (3) an additional 50% VOC emission reductions may not be effective in further reducing concentrations.     

Sampled Monte Carlo uncertainty analysis for photochemical grid models

This study reports on the development and testing of a method of quantifying the uncertainties in concentration predictions by a complex photochemical grid model (PGM), using a modification of the basic Monte Carlo method (MCM). The computationally intensive aspects of applying a full MCM to hundreds of PGM inputs and model parameters is replaced by a highly restricted sampling approach that exploits the spatial persistence found in predicted concentration fields. The sampling approach to the MCM is being explored as an efficient approach to assess the uncertainty in the differences in predicted maximum ozone concentration between base case and control scenarios. The MCM is applied to several dozen surface cells, with the goal of sampling the spatial pattern of uncertainty in the PGM-predicted differences in surface ozone concentration fields between a pair of base and control scenarios. The uncertainty in model inputs and parameters is simulated using several types of stochastic models. These stochastic models are driven using Latin hypercube sampling (LHS) to generate a non-redundant ensemble of alternative model inputs. Preliminary testing of the sampled MCM approach was conducted using the UAM-IV PGM on the New York ozone attainment modeling domain for the 6–8 July 1988 ozone episode. One hundred alternative concentration estimates were generated for a base scenario and for control scenarios representing 50%, 10% and 5% reduction of NO_x emissions. The upper and lower bounds of the concentration difference ensemble that define a 95% confidence range were spatially interpolated from 27 monitoring sites to the full (surface) modeling domain, using the field of zero uncertainty (ZU) concentration differences. For the 50% NO_x control scenario, predicted increases in peak ozone concentration smaller than 20 ppb were generally not significant from zero. By contrast, predicted decreases in peak ozone greater than 10 ppb were usually significant. For a control scenario with a small 5% NO_x reduction, predicted concentration differences and confidence intervals were much smaller, but predicted changes in peak ozone were significant at a number of sample cells.     

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.     

Meteorological Considerations in Air Quality Planning

Results from dispersion models which are routinely used for regulatory purposes do not reflect the uncertainties which are Inherent In the input data. To remedy this, a formula for propagating measurement uncertainties of emission rate, wind speed, wind direction, horizontal dispersion parameter, vertical dispersion parameter, effective emission height, and mixing depth is derived for EPA’s Industrial Source Complex Short Term (ISCST) Gaussian dispersion model for the simple case of a single stack-type source and nonbuoyant plume. Values for the uncertainties of the input variables are chosen and used to calculate ambient concentration uncertainties. These calculated uncertainties are compared with the standard deviation of ambient concentrations calculated from 2500 input data sets for each of four stability classes and three downwind distances, which were randomly altered to simulate the effects of measurement uncertainty. The calculated uncertainties do not differ significantly from the standard deviations of the randomized calculations for input data uncertainties as high as 30 percent and Stability Classes A-C. The calculated uncertainties overestimate the actual uncertainty of model calculations for input data uncertainties greater than 20 percent for Stability Class D.     

Robustness of modelled ozone exposures and doses

This study evaluates the robustness of the AOTX and AF(st)Y indices for assessing the ozone-induced risk to vegetation. These indices represent the accumulated concentration and stomatal flux, respectively, above a threshold value. The robustness is expressed as the sensitivity to changes in inputs and the uncertainty due to input errors. The input data are taken from a regional-scale chemical transport model. Both indices show increasing sensitivity with increasing threshold values. The sensitivity depends on the threshold and the characteristics of the frequency distribution for concentrations and stomatal fluxes. AF(st)Y appears less sensitive than AOTX for the thresholds adopted for critical levels. The couplings between concentration gradients and deposition algorithms complicate the assessment of the total uncertainty. For AF(st)Y, the uncertainty due to the modelled stomatal conductance may sometimes increase, but sometimes decrease, the overall uncertainty significantly. In particular, the maximum stomatal conductance plays an important role in determining the uncertainty.     

Global sensitivity analysis of a process-based model for ammonia emissions from manure storage and treatment structures

The development of process-based models to estimate ammonia emissions from animal feeding operations (AFOSs) is sought to replace costly and time-consuming direct measurements. Critical to process-based model development is conducting sensitivity analysis to determine the input parameters and their interactions that contribute most to the variance of the model output. Global and relative sensitivity analyses were applied to a process-based model for predicting ammonia emissions from the surface of anaerobic lagoons for treating and storing manure. The objectives were to compare global sensitivity analysis (GSA) to relative (local) sensitivity analysis (RSA) on a process-based model for ammonia emissions. Based on the first-order coefficient, both GSA and RSA showed the model input parameters in order of importance in process model for ammonia emissions from lagoon surfaces were: (i) pH, (ii) lagoon liquid temperature, (iii) wind speed above the lagoon surface, and (iv) the concentration of ammoniacal nitrogen in the lagoon. The GSA revealed that interactions between model parameters accounted for over two-thirds of the model variance, a result that cannot be achieved using traditional RSA. Also, the GSA showed that parameter interactions involving liquid pH had more impact on the model output variance than the single parameters: (i) temperature, (ii) wind speed, or (iii) total ammoniacal nitrogen. This study demonstrates that GSA provides a more complete analysis of model input parameters and their interactions on the model output compared to RSA. A comprehensive tutorial regarding the application of GSA to a process model is presented.     

An Atmospheric Dispersion Model for the Environmental Impact Assessment of Thermal Power Plants in Japan—A Method for Evaluating Topographical Effects

Abstract

An atmospheric dispersion model was developed for the environmental impact assessment of thermal power plants in Japan, and a method for evaluating topographical effects using this model was proposed. The atmospheric dispersion model consists of an airflow model with a turbulence closure model based on the algebraic Reynolds stress model and a Lagrangian particle dispersion model (LPDM). The evaluation of the maximum concentration of air pollutants such as SO₂, NO_x, and suspended particulate matter is usually considered of primary importance for environmental impact assessment. Three indices were therefore estimated by the atmospheric dispersion model: the ratios (α and β, respectively) of the maximum concentration and the distance of the point of the maximum concentration from the source over topography to the respective values over a flat plane, and the relative concentration distribution [γ(x)] along the ground surface projection of the plume axis normalized by the maximum concentration over a flat plane. The atmospheric dispersion model was applied to the topography around a power plant with a maximum elevation of more than 1000 m. The values of α and β evaluated by the atmospheric dispersion model varied between 1 and 3 and between 1 and 0.4, respectively, depending on the topographical features. These results and the calculated distributions of γ(x) were highly similar to the results of the wind tunnel experiment. Therefore, when the slope of a hill or mountain is similar to the topography considered in this study, it is possible to evaluate topographical effects on exhaust gas dispersion with reasonable accuracy using the atmospheric dispersion model as well as wind tunnel experiments.     

Control of Asphaltic Concrete Plants in Los Angeles County

Characteristics of maximum short-term ground level concentrations from an elevated point source, namely, the effective plume height, the critical wind speed, the distance to the point of maximum concentration, and the maximum concentration, are derived from the gaussian plume model. Both phases of plume development—before and after it has reached its final height—are considered. The plume rise treatment includes both thermal buoyancy and momentum effects. Certain limitations on critical wind speed are discussed. The dispersion model whose basis is established in this paper should be especially useful in applications where on site meteorological data are unavailable.     

A neural network-based approach for the prediction of urban SO2 concentrations in the Istanbul metropolitan area

A three-layer Artificial Neural Network (ANN) model was developed to forecast air pollution levels. The subsequent SO₂ concentration (24-hour averaged) being the output parameter of this study was estimated by seven input parameters such as preceding SO₂ concentrations (24-hour averaged), average daily temperature, sea-level pressure, relative humidity, cloudiness, average daily wind speed and daily dominant wind direction. After Backpropagation training combined with Principal Component Analysis (PCA), the proposed model predicted subsequent SO₂ values based on measured data. ANN testing outputs were proven to be satisfactory with correlation coefficients of about 0.770, 0.744 and 0.751 for the winter, summer and overall data, respectively.     

A multimedia environmental model of chemical distribution: fate, transport, and uncertainty analysis

This paper presented a framework for analysis of chemical concentration in the environment and evaluation of variance propagation within the model. This framework was illustrated through a case study of selected organic compounds of benzo[alpha]pyrene (BAP) and hexachlorobenzene (HCB) in the Great Lakes region. A multimedia environmental fate model was applied to perform stochastic simulations of chemical concentrations in various media. Both uncertainty in chemical properties and variability in hydrometeorological parameters were included in the Monte Carlo simulation, resulting in a distribution of concentrations in each medium. Parameters of compartmental dimensions, densities, emissions, and background concentrations were assumed to be constant in this study. The predicted concentrations in air, surface water and sediment were compared to reported data for validation purpose. Based on rank correlations, a sensitivity analysis was conducted to determine the influence of individual input parameters on the output variance for concentration in each environmental medium and for the basin-wide total mass inventory. Results of model validation indicated that the model predictions were in reasonable agreement with spatial distribution patterns, among the five lake basins, of reported data in the literature. For the chemical and environmental parameters given in this study, parameters associated to air-ground partitioning (such as moisture in surface soil, vapor pressure, and deposition velocity) and chemical distribution in soil solid (such as organic carbon partition coefficient and organic carbon content in root-zone soil) were targeted to reduce the uncertainty in basin-wide mass inventory. This results of sensitivity analysis in this study also indicated that the model sensitivity to an input parameter might be affected by the magnitudes of input parameters defined by the parameter settings in the simulation scenario. Therefore, uncertainty and sensitivity analyses for environmental fate models was suggested to be conducted after the model output was validated based on an appropriate input parameter settings.     

An atmospheric dispersion model for the environmental impact assessment of thermal power plants in Japan--a method for evaluating topographical effects

An atmospheric dispersion model was developed for the environmental impact assessment of thermal power plants in Japan, and a method for evaluating topographical effects using this model was proposed. The atmospheric dispersion model consists of an airflow model with a turbulence closure model based on the algebraic Reynolds stress model and a Lagrangian particle dispersion model (LPDM). The evaluation of the maximum concentration of air pollutants such as SO2, NOx, and suspended particulate matter is usually considered of primary importance for environmental impact assessment. Three indices were therefore estimated by the atmospheric dispersion model: the ratios (alpha and beta, respectively) of the maximum concentration and the distance of the point of the maximum concentration from the source over topography to the respective values over a flat plane, and the relative concentration distribution [gamma(x)] along the ground surface projection of the plume axis normalized by the maximum concentration over a flat plane. The atmospheric dispersion model was applied to the topography around a power plant with a maximum elevation of more than 1,000 m. The values of alpha and beta evaluated by the atmospheric dispersion model varied between 1 and 3 and between 1 and 0.4, respectively, depending on the topographical features. These results and the calculated distributions of y(x) were highly similar to the results of the wind tunnel experiment. Therefore, when the slope of a hill or mountain is similar to the topography considered in this study, it is possible to evaluate topographical effects on exhaust gas dispersion with reasonable accuracy using the atmospheric dispersion model as well as wind tunnel experiments.     

Application of HYDRUS 1D model for assessment of phenol–soil adsorption dynamics

Laboratory-scale batch, vertical, and horizontal column experiments were conducted to investigate the attenuative capacity of a fine-grained clayey soil of local origin in the surrounding of a steel plant wastewater discharge site in West Bengal, India, for removal of phenol. Linear, Langmuir, and Freundlich isotherm plots from batch experimental data revealed that Freundlich isotherm model was reasonably fitted (R ²?=?0.94). The breakthrough column experiments were also carried out with different soil bed heights (5, 10, and 15 cm) under uniform flow to study the hydraulic movements of phenol by evaluating time concentration flow behavior using bromide as a tracer. The horizontal migration test was also conducted in the laboratory using adsorptive phenol and nonreactive bromide tracer to explore the movement of solute in a horizontal distance. The hydrodynamic dispersion coefficients (D) in the vertical and horizontal directions in the soil were estimated using nonlinear least-square parameter optimization method in CXTFIT model. In addition, the equilibrium convection dispersion model in HYDRUS 1D was also examined to simulate the fate and transport of phenol in vertical and horizontal directions using Freundlich isotherm constants and estimated hydrodynamic parameters as input in the model. The model efficacy and validation were examined through statistical parameters such as the coefficient of determination (R ²), root mean square error and design of index (d).     

On the uncertainty and confidence in decision support tools (DSTs) with insights from the Baltic Sea ecosystem

Ecosystems around the world are increasingly exposed to multiple, often interacting human activities, leading to pressures and possibly environmental state changes. Decision support tools (DSTs) can assist environmental managers and policy makers to evaluate the current status of ecosystems (i.e. assessment tools) and the consequences of alternative policies or management scenarios (i.e. planning tools) to make the best possible decision based on prevailing knowledge and uncertainties. However, to be confident in DST outcomes it is imperative that known sources of uncertainty such as sampling and measurement error, model structure, and parameter use are quantified, documented, and addressed throughout the DST set-up, calibration, and validation processes. Here we provide a brief overview of the main sources of uncertainty and methods currently available to quantify uncertainty in DST input and output. We then review 42 existing DSTs that were designed to manage anthropogenic pressures in the Baltic Sea to summarise how and what sources of uncertainties were addressed within planning and assessment tools. Based on our findings, we recommend future DST development to adhere to good modelling practise principles, and to better document and communicate uncertainty among stakeholders.Electronic supplementary materialThe online version of this article (doi:10.1007/s13280-020-01385-x) contains supplementary material, which is available to authorized users.     

A sensitivity analysis of adsorption and degradation parameters in the modeling of pesticide transport in soils

The increasing use of deterministic models in predicting the movement of pesticides in soils, has focused attention on the evaluation of major parameters which represent attenuation factors of organics in the subsurface. These parameters are the degradation rate constant and the adsorption constant for the pesticide. In view of the large in situ variability of these parameters and of the difficulty in obtaining accurate field data, there is a high degree of uncertainty associated with the results obtained from deterministic models. A sensitivity analysis is performed here to quantify the impact of such variation in each of these input parameters on the output results of an unsaturated zone transport model (PRZM). Results show that variations in these parameters about their respective mean values greatly affect the predicted concentration distributions, obtained after three years, of the pesticide aldicarb in all the soil profile. A 15–22% variation in the degradation constant, or a 24% variation in the adsorption constant, lead to a 100% uncertainty in the various simulation results defined as the cumulative quantity of aldicarb or the dissolved aldicarb concentration leached below the root zone (or the unsaturated zone) of the soil. Such a deterministic model presents a high degree of sensitivity to these input parameters. Accurate field data are then needed to obtain reliable model results in predicting pesticide movement inthe unsaturated zone.