Testing the Potential for Predictive Modeling and Mapping and Extending Its Use as a Tool for Evaluating Management Scenarios and Economic Valuation in the Baltic Sea (PREHAB)

We evaluated performance of species distribution models for predictive mapping, and how models can be used to integrate human pressures into ecological and economic assessments. A selection of 77 biological variables (species, groups of species, and measures of biodiversity) across the Baltic Sea were modeled. Differences among methods, areas, predictor, and response variables were evaluated. Several methods successfully predicted abundance and occurrence of vegetation, invertebrates, fish, and functional aspects of biodiversity. Depth and substrate were among the most important predictors. Models incorporating water clarity were used to predict increasing cover of the brown alga bladderwrack Fucus vesiculosus and increasing reproduction area of perch Perca fluviatilis, but decreasing reproduction areas for pikeperch Sander lucioperca following successful implementation of the Baltic Sea Action Plan. Despite variability in estimated non-market benefits among countries, such changes were highly valued by citizens in the three Baltic countries investigated. We conclude that predictive models are powerful and useful tools for science-based management of the Baltic Sea.     

Selection of an appropriate model to predict plume dispersion in coastal areas

In order to suggest a new methodology for selecting an appropriate dispersion model, various statistical measures having respective characteristics and recommended value ranges were integrated to produce a new single index by using fuzzy inference where eight statistical measures for various model results, including fractional bias (FB), normalized mean square error (NMSE), geometric bias mean (MG), geometric bias variance (VG), within a factor of two (FAC2), index of agreement (IOA), unpaired accuracy of the peak concentration (UAPC), and mean relative error (MRE), were taken as premise part variables. The new methodology using a single index was applied to the prediction of ground-level SO₂ concentration of 1-h average in coastal areas, where eight modeling combinations were organized with fumigation models, σ_y schemes for pre-fumigation, and modification schemes for σ_y during fumigation. As a result, the fumigation model of Lyons and Cole was found to have better predictability than the modified Gaussian model assuming that whole plume is immerged into the Thermal Internal Boundary Layer (TIBL). Again, a better scheme of σ_y (fumigation) was discerned. This approach, which employed the new integrated index, appears to be applicable to model evaluation or selection in various areas including complex coastal areas.     

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.     

Comparison of land-use regression models between Great Britain and the Netherlands

Land-use regression models have increasingly been applied for air pollution mapping at typically the city level. Though models generally predict spatial variability well, the structure of models differs widely between studies. The observed differences in the models may be due to artefacts of data and methodology or underlying differences in source or dispersion characteristics. If the former, more standardised methods using common data sets could be beneficial. We compared land-use regression models for NO₂ and PM_10, developed with a consistent protocol in Great Britain (GB) and the Netherlands (NL).Models were constructed on the basis of 2001 annual mean concentrations from the national air quality networks. Predictor variables used for modelling related to traffic, population, land use and topography. Four sets of models were developed for each country. First, predictor variables derived from data sets common to both countries were used in a pooled analysis, including an indicator for country and interaction terms between country and the identified predictor variables. Second, the common data sets were used to develop individual baseline models for each country. Third, the country-specific baseline models were applied after calibration in the other country to explore transferability. The fourth model was developed using the best possible predictor variables for each country.A common model for GB and NL explained NO₂ concentrations well (adjusted R² 0.64), with no significant differences in intercept and slopes between the two countries. The country-specific model developed on common variables for NL but not GB improved the prediction.The performance of models based upon common data was only slightly worse than models optimised with local data. Models transferred to the other country performed substantially worse than the country-specific models. In conclusion, care is needed both in transferring models across different study areas, and in developing large inter-regional LUR models.     

Simple PDF models for convectively driven vertical diffusion

The mode of vertical velocity in convective boundary layers (CBLs) is usually negative and the probability distribution function (PDF) of w, P_w, is rarely symmetric except near the top and bottom of CBLs. Consequently, vertical diffusion from elevated sources is usually asymmetric and exhibits a descending mode of concentration, causing higher peak surface concentrations than predicted by Gaussian models. The main concentration (χ) effects, we argue, can be modeled using the simplest of PDF diffusion models, with tracers responding to P_w at the source height with straight line trajectories and simple reflection at the surface and z_i, the mixing depth. The critical element is the choice of P_w. Two P_w models are offered, a bi-Gaussian (BG) and a Gaussian-ramp (GR) formulation. Both have some observational support, and the resulting PDF models are mathematically tractable. Analytical solutions for key variables are given; these show some surprising contrasts between the BG and GR models, but both can approximate laboratory and numerical modeling results for ∝χdy patterns. A diverse selection of atmospheric turbulence measurements is presented; for measures that reflect asymmetry in P_w, the data show wide ranges and do not lend support to any particular form of P_w. Recent lidar measurements of oil fog plumes are presented that show a large variability in ∝χdy patterns, even with substantial averaging periods. The only concurrent turbulence measurement that strongly correlates with the observed vertical diffusion of oil fog is the mode of wind elevation angle. A simple adaptation of the BG model is recommended that fits the average peak ∝χdy and distance of occurrence as observed so far.     

Weighted road density: A simple way of assigning traffic-related air pollution exposure

BackgroundExisting traffic variables used for predicting NO₂ in epidemiological studies are either difficult to acquire or explain only a small proportion of the variance. The purpose of this study was to develop and validate a new predictor, weighted road density, which combines the maximum amount of information related to traffic into a single variable without the requirement of obtaining traffic counts for a given area.MethodTwo week NO₂ samples were collected using the readings of up to 32 passive samplers on 3 separate rounds between September and December 2006 and again in 2007. Several types of traffic related explanatory variables based on traffic counts, distance to main road and the proposed weighted road density were constructed using GIS software, and tested for association with the NO₂ samplers. Assessment of the best model was based on R² values, as well as leave-one-out cross validation.ResultsThe weighted road density variable and the density variable based on traffic counts resulted in a similar R² (0.59) for predicting NO₂, although weighted road density was much easier to construct and outperformed other variables such as distance to main road.ConclusionAs well as being a powerful predictor for use in a land use regression model, weighted road density can be used as a proxy for exposure to traffic-related pollution, for use in circumstances where direct measurements of pollutant levels are not feasible or are not required.     

Estimated long-term outdoor air pollution concentrations in a cohort study

Several recent studies associated long-term exposure to air pollution with increased mortality. An ongoing cohort study, the Netherlands Cohort Study on Diet and Cancer (NLCS), was used to study the association between long-term exposure to traffic-related air pollution and mortality. Following on a previous exposure assessment study in the NLCS, we improved the exposure assessment methods.Long-term exposure to nitrogen dioxide (NO₂), nitrogen oxide (NO), black smoke (BS), and sulphur dioxide (SO₂) was estimated. Exposure at each home address (N=21 868) was considered as a function of a regional, an urban and a local component. The regional component was estimated using inverse distance weighed interpolation of measurement data from regional background sites in a national monitoring network. Regression models with urban concentrations as dependent variables, and number of inhabitants in different buffers and land use variables, derived with a Geographic Information System (GIS), as predictor variables were used to estimate the urban component. The local component was assessed using a GIS and a digital road network with linked traffic intensities. Traffic intensity on the nearest road and on the nearest major road, and the sum of traffic intensity in a buffer of 100 m around each home address were assessed. Further, a quantitative estimate of the local component was estimated.The regression models to estimate the urban component explained 67%, 46%, 49% and 35% of the variances of NO₂, NO, BS, and SO₂ concentrations, respectively. Overall regression models which incorporated the regional, urban and local component explained 84%, 44%, 59% and 56% of the variability in concentrations for NO₂, NO, BS and SO₂, respectively.We were able to develop an exposure assessment model using GIS methods and traffic intensities that explained a large part of the variations in outdoor air pollution concentrations.     

A land use regression for predicting fine particulate matter concentrations in the New York City region

We developed regression equations to predict fine particulate matter (PM_2.5) at air monitoring locations in the New York City region using data on nearby traffic and land use patterns. Three-year averages (1999–2001) of PM_2.5 at US Environmental Protection Agency (EPA) monitors in the 28 counties including and surrounding New York City were calculated using daily data from the EPA's Air Quality Subsystem. As the secondary contribution to PM_2.5 concentrations is lowest in the winter, we also calculated and modeled average winter 2000 PM_2.5 to conduct a preliminary evaluation of model sensitivity to source contribution. Candidate predictor variables included traffic, land use, census and emissions data from local, state and national sources and were tabulated for a series of circular buffer regions at varying distances around the monitors using a geographic information system. In total, more than 25 variables at 5 different buffer distances were considered for inclusion in the model. Before evaluating the variables we removed several samples from the modeling for validation. For comparison and validation purposes we computed both a model using data for the full 28-county region as well as a more urbanized 9-county region. We found that traffic within a buffer of 300 or 500 m explains the greatest proportion of variance (37–44%) in all 3 models. Measures of urbanization, specifically population density, explain a significant amount of the residual variation (7–18%) after including a traffic variable. Finally, a measure of industrial land use further improves the 28-county and 9-county models based on the 3-yr annual averages, explaining an additional 4% and 11% of the variation, respectively, while vegetative land use improves the winter model explaining an additional 6%. The final models predicted well at validation locations. In total, the final land use regression models explain between 61% and 64% of the variation in PM_2.5.     

Comparison of the performances of land use regression modelling and dispersion modelling in estimating small-scale variations in long-term air pollution concentrations in a Dutch urban area

The performance of a Land Use Regression (LUR) model and a dispersion model (URBIS – URBis Information System) was compared in a Dutch urban area. For the Rijnmond area, i.e. Rotterdam and surroundings, nitrogen dioxide (NO₂) concentrations for 2001 were estimated for nearly 70 000 centroids of a regular grid of 100 × 100 m.A LUR model based upon measurements carried out on 44 sites from the Dutch national monitoring network and upon Geographic Information System (GIS) predictor variables including traffic intensity, industry, population and residential land use was developed. Interpolation of regional background concentration measurements was used to obtain the regional background. The URBIS system was used to estimate NO₂ concentrations using dispersion modelling. URBIS includes the CAR model (Calculation of Air pollution from Road traffic) to calculate concentrations of air pollutants near urban roads and Gaussian plume models to calculate air pollution levels near motorways and industrial sources. Background concentrations were accounted for using 1 × 1 km maps derived from monitoring and model calculations.Moderate agreement was found between the URBIS and LUR in calculating NO₂ concentrations (R = 0.55). The predictions agreed well for the central part of the concentration distribution but differed substantially for the highest and lowest concentrations. The URBIS dispersion model performed better than the LUR model (R = 0.77 versus R = 0.47 respectively) in the comparison between measured and calculated concentrations on 18 validation sites. Differences can be understood because of the use of different regional background concentrations, inclusion of rather coarse land use category industry as a predictor variable in the LUR model and different treatment of conversion of NO to NO₂.Moderate agreement was found between a dispersion model and a land use regression model in calculating annual average NO₂ concentrations in an area with multiple sources. The dispersion model explained concentrations at validation sites better.     

Modeling the intra-urban variability of outdoor traffic pollution in Oslo,Norway—A GA2LEN project

Traffic is a major source of air pollutants in urban environments, and exposure to these pollutants may be associated with adverse health effects. However, inconsistencies in observational epidemiological studies may be caused by differential measurement errors in various approaches in assessing exposure.We aimed to evaluate a simple method for assessing outdoor air pollutant concentrations in Oslo, Norway, through a land-use regression method.Samples of nitrogen oxides (NO_x) were collected in two different weeks using Ogawa passive diffusion samplers simultaneously at 80 locations across Oslo. Independent variables used in subsequent regression models as predictors of the pollutants were derived using the Arc 9 geographic information system (GIS) software. Indicators of land use, traffic, population density, and physical geography were tested.The final regression model yielded an adjusted coefficient of determination (R²) of 0.77 for nitrogen dioxide (NO₂), 0.66 for nitric oxide (NO), and 0.73 for NO_x.The results suggest that a good predictive exposure model can be derived from this approach, which can be used to estimate long-term small-area variation in concentrations for individual exposure assessment in epidemiological studies in a highly cost-effective way. These small-area variations in traffic pollution are important since they may have associations with health effects.     

Analysis of model responses to emission-reduction scenarios within the CityDelta project

This paper investigates how air quality models applied at different scales (50 and 5 km horizontal resolutions) can predict pollution levels in response to emission control strategies in various cities in Europe. This study, involving five modelling teams and focused on four European cities, has been conducted within the CityDelta project (http://aqm.jrc.it/citydelta). The CityDelta models generally agree, on the O₃ changes expected from scenarios representative of the current legislation on air pollution in 2010, named CLE. They also agree about less scope for further improvements from emission controls beyond CLE. For PM₁₀, more significant differences between the models are observed, especially between models with different spatial resolutions. However, these differences are city-dependent and are larger in complex geographical areas such as Milan in the Pô Valley than in the Paris area.Fine scale models generally capture important urban scale effects, which are not represented by regional scale models. For instance, they improve the simulation of potential O₃ increase caused by NO_x emissions reduction in NMVOC-limited regime situations. Large scale models generally underpredict PM mean concentrations in city areas. A series of emission scenarios to address the question of the efficiency of local emission controls designed independently from regional measures is analyzed. The analysis of the CityDelta results contributes to the quantification of the impact of grid resolution in air quality modelling, and its application to emission control scenarios.     

Assessing the spatial distribution of nitrogen dioxide in London,Ontario

Land use regression (LUR) models have been widely used to characterize the spatial distribution of urban air pollution and estimate exposure in epidemiologic studies. However, spatial patterns of air pollution vary greatly between cities due to local source type and distribution. London, Ontario, Canada, is a medium-sized city with relatively few and isolated industrial point sources, which allowed the study to focus on the contribution of different transportation sectors to urban air pollution. This study used LUR models to estimate the spatial distribution of nitrogen dioxide (NO₂) and to identify local sources influencing NO₂ concentrations in London, ON. Passive air sampling was conducted at 50 locations throughout London over a 2-week period in May–June 2010. NO₂ concentrations at the monitored locations ranged from 2.8 to 8.9 ppb, with a median of 5.2 ppb. Industrial land use, dwelling density, distance to highway, traffic density, and length of railways were significant predictors of NO₂ concentrations in the final LUR model, which explained 78% of NO₂ variability in London. Traffic and dwelling density explained most of the variation in NO₂ concentrations, which is consistent with LUR models developed in other Canadian cities. We also observed the importance of local characteristics. Specifically, 17% of the variation was explained by distance to highways, which included the impacts of heavily traveled corridors transecting the southern periphery of the city. Two large railway yards and railway lines throughout central areas of the city explained 9% of NO₂ variability. These results confirm the importance of traditional LUR variables and highlight the importance of including a broader array of local sources in LUR modeling. Finally, future analyses will use the model developed in this study to investigate the association between ambient air pollution and cardiovascular disease outcomes, including plaque burden, cholesterol, and hypertension.

Implications: Monitoring and modeling of NO₂ throughout the city of London represents an important step toward assessing air pollution health effects in a mid-sized Canadian city. The study supports the introduction of railways to LUR modeling of NO₂. Railways explained approximately 9% of the variability in ambient NO₂ concentrations in London, which suggests that local sources captured by land-use indicators may contribute to the efficacy of LUR models. These findings provide insights relevant to other medium and smaller sized cities with similar land use and transportation infrastructure. Furthermore, London is a central hub for medical research and treatment in southwestern Ontario, with facilities such as the Robarts Research Institute, London Regional Cancer Program (LRCP), and Stroke Prevention & Atherosclerosis Research Centre (SPARC). The models developed in this study will provide estimates of exposure for future analyses examining air pollution health effects in this data-rich population.     

Modeling the non-point source pollution risks by combing pollutant sources,precipitation, and landscape structure

Traditional models of nutrient simulation usually focus on the pollutant sources and precipitation, lacking the quantification of landscape structure. We developed a new prediction model of pollution risks by combing pollutant sources, precipitation, and landscape structure, which was defined as the source-precipitation-landscape model (SPLM). The SPLM was applied to simulate the non-point source (NPS) total nitrogen (TN) exports in one of the largest river basins in China (the Haihe River Basin, HRB). TN concentrations of 35 sampling catchments in 2013 were used to test the accuracy of the SPLM. Simulated results showed that (1) the SPLM had a relative high accuracy in the simulation of NPS TN export and intensity, especially for TN intensity. (2) The mean TN export and intensity of all the 1578 catchments in the HRB were 441.97 t and 2.08 t/km², respectively. (3) The TN export intensities differed greatly among the sub-basins in the HRB, ranging from 0.64 to 6.81 t/km². On the whole, the TN export intensities of the plain sub-basins (e.g., the Tuhaimajia River, the Heilonggang River, and the Beisihe River) were much higher than those of mountainous sub-basins (e.g., the Yongding River, the Beisanhe River, and the Luanhe River). (4) The contributions to TN exports, from high to low, were land use (38.82%), livestock husbandry (33.57%), and rural population (27.61%). Among all the ten pollution sources, arable land (30.87%), rural population (27.61%), and large livestock (17.73%) had the top three contributions to TN exports. This study provides a feasible tool for policymakers and administrators to develop workable management measures for the mitigation of NPS pollution. This SPLM can be extended to other regions in a rapid urbanization context.

     

Numerical simulation of dispersion around an isolated cubic building: Comparison of various types of k–ɛ models

Prediction accuracy of flow and dispersion around a cubic building with a flush vent located on its roof was examined using various k–? models, and numerical results were compared with wind-tunnel data. Four types of turbulence models, i.e., the standard k–? model, the RNG k–? model, the k–? model with Launder and Kato modification and the Realizable k–? model were compared in this study. The standard k–? model provided inadequate results for the concentration field, because it could not reproduce the basic flow structure, such as the reverse flow on the roof. However, revised k–? models provided concentrations in better agreement with the experimental data. The effect of an oblique wind angle and vent locations on the prediction accuracy was also investigated. It was confirmed that the prediction accuracy of the velocity field strongly affected that of the concentration field. The RNG model showed general agreement with the experiment, and was the best of the turbulence models tested. However, it becomes clear that the results for all CFD models show poor prediction accuracy of concentration distribution at the side and leeward surfaces of the building since they all underestimate the concentration diffusion on these regions. The concentrations predicted by all CFD models were less diffusive than those of the experiment.     

Prediction of dissolved oxygen concentration in hypoxic river systems using support vector machine: a case study of Wen-Rui Tang River,China

Accurate quantification of dissolved oxygen (DO) is critically important for managing water resources and controlling pollution. Artificial intelligence (AI) models have been successfully applied for modeling DO content in aquatic ecosystems with limited data. However, the efficacy of these AI models in predicting DO levels in the hypoxic river systems having multiple pollution sources and complicated pollutants behaviors is unclear. Given this dilemma, we developed a promising AI model, known as support vector machine (SVM), to predict the DO concentration in a hypoxic river in southeastern China. Four different calibration models, specifically, multiple linear regression, back propagation neural network, general regression neural network, and SVM, were established, and their prediction accuracy was systemically investigated and compared. A total of 11 hydro-chemical variables were used as model inputs. These variables were measured bimonthly at eight sampling sites along the rural-suburban-urban portion of Wen-Rui Tang River from 2004 to 2008. The performances of the established models were assessed through the mean square error (MSE), determination coefficient (R ²), and Nash-Sutcliffe (NS) model efficiency. The results indicated that the SVM model was superior to other models in predicting DO concentration in Wen-Rui Tang River. For SVM, the MSE, R ², and NS values for the testing subset were 0.9416 mg/L, 0.8646, and 0.8763, respectively. Sensitivity analysis showed that ammonium-nitrogen was the most significant input variable of the proposal SVM model. Overall, these results demonstrated that the proposed SVM model can efficiently predict water quality, especially for highly impaired and hypoxic river systems.     

Evaluation of the distribution of fecal indicator bacteria in a river system depending on different types of land use in the southern watershed of the Baltic Sea

The aim of the study was to determine the effects of land use management on changes in the fecal contamination of water in the ?yna River, one of the main lowland watercourses in the southern watershed of the Baltic Sea (northern Poland). A total of 120 water samples were collected in different seasons of 2011 and 2012 at 15 sites where the river intersected forest (FA), agricultural (AA), and urbanized (UA) areas. Fecal indicator bacteria (FIB), the counts of Enterobacteriaceae and Escherichia coli, total bacterial counts (TBCs), and domain Bacteria (EUB338) were determined by culture-dependent and culture-independent methods. Temperature, pH, chemical oxygen demand, dissolved oxygen, total dissolved solids, ammonia nitrogen, nitrite nitrogen, nitrate nitrogen, orthophosphate, and total phosphorus were also determined. The lowest bacterial counts were noted in water samples collected in FA, and the highest in samples collected in UA. Statistically significant differences were determined between bacterial populations across the analyzed land use types and in different sampling seasons. Significant correlations were also observed between the populations of FIB and physicochemical parameters. The results indicate that land use type influenced FIB concentrations in river water. The combined use of conventional and molecular methods improves the accuracy of fecal contamination analyses in river ecosystems.     

Simulation of the daily average PM10 concentrations at Ta-Liao with Box–Jenkins time series models and multivariate analysis

Southern Taiwan has experienced severe PM₁₀ problems for over a decade. The present paper describes the establishment of a simulation model for the daily average PM₁₀ concentrations at Ta-Liao, southern Taiwan. The study used a regression with time series error models (RTSE models) (multivariate ARIMA time series model), including an explanatory variable resulting from principal component analyses to complete the PM₁₀ simulation. Factor 1 estimated from the factor analyses explained the variance of 44–49%, which indicated the important contribution from the neighbor-city PM₁₀ at Mei-Nung, Lin-Yuang, Zuoying, Chao-Chou, local ozone and NOx. Factor 1 can be interpreted with regional PM₁₀ plus photochemical reactions. To improve the predictability of extremely high PM₁₀, different results from the principal component analysis were introduced to the RTSE models. We constructed four kinds of RTSE models: RTSE model without PC, with PC4S (PM₁₀ at Mei-Nung, Lin-Yuang, Zuoying, and Chao-Chou), with PCTL (meteorological variables and co-pollutants at Ta-Liao), and with PCTL4S (the combination of the above two) and evaluated the statistics model performance. Ozone, dew point temperature, NOx, wind speed, wind directions, and the PC trigger were the significant variables in the RTSE models most of time. When the neighbor-city PM₁₀ was included in the PC trigger, the predictability was apparently improved. The closeness of fit with the inclusion of PC4S and PCTL4S was improved by reducing SEE from 0.117 to 0.092. Using the RTSE models with PC4S or PCTL4S, POD was improved by an increase of 33%, FAR was reduced 30%, and CSI was increased 39%, when simulating the daily average PM₁₀ > 150 μg m^?3. Evidently we need to survey source impacts prior to establishing a simulation model. Factor analysis is a useful method to investigate sources that contributed PM₁₀ to a target site prior to establishing a simulation model.     

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.     

The emergence of land systems as the nexus for sustainability transformations: This article belongs to Ambio's 50th Anniversary Collection. Theme: Agriculture land use

This perspective recognizes the seminal Ambio articles of Sombroek et al. (1993), Turner et al. (1994) and Brussaard et al. (1997), identifying their individual and collective role in laying the ground work for a global change research agenda on land and its human use through increased understanding of terrestrial ecosystem dynamics and global change, and furthering nascent interdisciplinary efforts within the global change science community to better understand the ‘human driving forces’ of change. From these efforts, land system science, as a systemic science focused on complex socio-ecological interactions around land use and associated trade-offs and synergies, emerges as an ‘interdiscipline’ challenged to better understand land systems as the ‘meeting ground’ for multiple claims on land for biodiversity, carbon, livelihoods, food production among others, and support pathways to sustainability for people and nature.     

Extreme learning machines: a new approach for modeling dissolved oxygen (DO) concentration with and without water quality variables as predictors

In this paper, several extreme learning machine (ELM) models, including standard extreme learning machine with sigmoid activation function (S-ELM), extreme learning machine with radial basis activation function (R-ELM), online sequential extreme learning machine (OS-ELM), and optimally pruned extreme learning machine (OP-ELM), are newly applied for predicting dissolved oxygen concentration with and without water quality variables as predictors. Firstly, using data from eight United States Geological Survey (USGS) stations located in different rivers basins, USA, the S-ELM, R-ELM, OS-ELM, and OP-ELM were compared against the measured dissolved oxygen (DO) using four water quality variables, water temperature, specific conductance, turbidity, and pH, as predictors. For each station, we used data measured at an hourly time step for a period of 4 years. The dataset was divided into a training set (70%) and a validation set (30%). We selected several combinations of the water quality variables as inputs for each ELM model and six different scenarios were compared. Secondly, an attempt was made to predict DO concentration without water quality variables. To achieve this goal, we used the year numbers, 2008, 2009, etc., month numbers from (1) to (12), day numbers from (1) to (31) and hour numbers from (00:00) to (24:00) as predictors. Thirdly, the best ELM models were trained using validation dataset and tested with the training dataset. The performances of the four ELM models were evaluated using four statistical indices: the coefficient of correlation (R), the Nash-Sutcliffe efficiency (NSE), the root mean squared error (RMSE), and the mean absolute error (MAE). Results obtained from the eight stations indicated that: (i) the best results were obtained by the S-ELM, R-ELM, OS-ELM, and OP-ELM models having four water quality variables as predictors; (ii) out of eight stations, the OP-ELM performed better than the other three ELM models at seven stations while the R-ELM performed the best at one station. The OS-ELM models performed the worst and provided the lowest accuracy; (iii) for predicting DO without water quality variables, the R-ELM performed the best at seven stations followed by the S-ELM in the second place and the OP-ELM performed the worst with low accuracy; (iv) for the final application where training ELM models with validation dataset and testing with training dataset, the OP-ELM provided the best accuracy using water quality variables and the R-ELM performed the best at all eight stations without water quality variables. Fourthly, and finally, we compared the results obtained from different ELM models with those obtained using multiple linear regression (MLR) and multilayer perceptron neural network (MLPNN). Results obtained using MLPNN and MLR models reveal that: (i) using water quality variables as predictors, the MLR performed the worst and provided the lowest accuracy in all stations; (ii) MLPNN was ranked in the second place at two stations, in the third place at four stations, and finally, in the fourth place at two stations, (iii) for predicting DO without water quality variables, MLPNN is ranked in the second place at five stations, and ranked in the third, fourth, and fifth places in the remaining three stations, while MLR was ranked in the last place with very low accuracy at all stations. Overall, the results suggest that the ELM is more effective than the MLPNN and MLR for modelling DO concentration in river ecosystems.