Forecasting air quality time series using deep learning

This paper presents one of the first applications of deep learning (DL) techniques to predict air pollution time series. Air quality management relies extensively on time series data captured at air monitoring stations as the basis of identifying population exposure to airborne pollutants and determining compliance with local ambient air standards. In this paper, 8 hr averaged surface ozone (O₃) concentrations were predicted using deep learning consisting of a recurrent neural network (RNN) with long short-term memory (LSTM). Hourly air quality and meteorological data were used to train and forecast values up to 72 hours with low error rates. The LSTM was able to forecast the duration of continuous O₃ exceedances as well. Prior to training the network, the dataset was reviewed for missing data and outliers. Missing data were imputed using a novel technique that averaged gaps less than eight time steps with incremental steps based on first-order differences of neighboring time periods. Data were then used to train decision trees to evaluate input feature importance over different time prediction horizons. The number of features used to train the LSTM model was reduced from 25 features to 5 features, resulting in improved accuracy as measured by Mean Absolute Error (MAE). Parameter sensitivity analysis identified look-back nodes associated with the RNN proved to be a significant source of error if not aligned with the prediction horizon. Overall, MAE's less than 2 were calculated for predictions out to 72 hours.

Implications: Novel deep learning techniques were used to train an 8-hour averaged ozone forecast model. Missing data and outliers within the captured data set were replaced using a new imputation method that generated calculated values closer to the expected value based on the time and season. Decision trees were used to identify input variables with the greatest importance. The methods presented in this paper allow air managers to forecast long range air pollution concentration while only monitoring key parameters and without transforming the data set in its entirety, thus allowing real time inputs and continuous prediction.     

AQI time series prediction based on a hybrid data decomposition and echo state networks

Environmental Science and Pollution Research - A hybrid AQI time series prediction model is proposed based on EWT-SE-VMD secondary decomposition, ICA (imperialist competitive algorithm) feature...     

An improved trend vegetation analysis for non-stationary NDVI time series based on wavelet transform

Environmental Science and Pollution Research - The aim of this paper is to improve trend analysis for non-stationary Normalized Difference Vegetation Index (NDVI) time series (TS) over different...     

Environmental monitoring model for a drainage basin obtained through spectral analysis of time series

The quality of results of an environmental monitoring plan is limited to the weakest component, which could be the analytical approach or sampling method. Considering both the possibilities and the fragility that sampling methods offer, this environmental monitoring study focused on the uncertainties caused by the time component. Four time series of nutrient concentration at two sampling points (PB1 and PB2) in the Ribeir?o Garcia basin in Blumenau, Brazil, which were significantly correlated to the spatial component, were considered with a 2-hour resolution to develop efficient sampling models. These models were based on the time at which there was the highest tendency toward adverse environmental effects. Fourier spectral analysis was used to evaluated the time series and resulted in two sampling models: (1) the SMCP (sampling model for critical period) that operated with 100% efficiency for registering the highest concentration of nutrients and was valid for 83% of the studied parameters; and (2) the SMGCP (sampling model for global critical period) that operated with 83 and 50% efficiency for PB1 and PB2, respectively.     

Autocorrelation and univariate time series modelling for grass fluoride and airborne fluoride concentrations

The dependence of observed fluoride levels-grass, gaseous air and particulate air-on previous levels was investigated in the field situation. Autocorrelation was found in grass fluoride observations, and to a lesser extent was present in both gaseous and particulate airborne fluorides. Univariate time series models were obtained which accounted for 56-66% of total variation in grass fluoride, 31% in gaseous airborne and 26% in particulate airborne fluorides. The large amount of variation unexplained by the models was thought to be due to the influence of environmental and meteorological factors not included in the models, and random variation due to day-to-day and plot-to-plot variation.     

Unsupervised monitoring of vegetation in a surface coal mining region based on NDVI time series

Environmental Science and Pollution Research - Surface coal mining causes vegetation disturbance while providing an energy source. Thus, much attention is given to monitoring the vegetation of...     

Risk assessment and validation of farmland abandonment based on time series change detection

Environmental Science and Pollution Research - Farmland abandonment, a widespread phenomenon during land-use transition, leads to a cycling or vanishing evolution of farmland resources. As...     

Autoregressive models in environmental forecasting time series: a theoretical and application review

Environmental Science and Pollution Research - Though globalization, industrialization, and urbanization have escalated the economic growth of nations, these activities have played foul on the...     

Weather integrated malaria prediction system using Bayesian structural time series model for northeast states of India

Environmental Science and Pollution Research - Malaria is an endemic disease in India and targeted to eliminate by the year 2030. The present study is aimed at understanding the epidemiological...     

An investigation of distributed lag models in the context of air pollution and mortality time series analysis

In particulate air pollution mortality time series studies, the particulate air pollution exposure measure used is typically the current day's or the previous day's air pollution concentration or a multi-day moving average air pollution concentration. Distributed lag models (DLMs) that allow for differential air pollution effects that are spread over multiple days are seen as an improvement over using a single- or multi-day moving average air pollution exposure measure. However, at the current time, the statistical properties of DLMs as a measure of air pollution exposure have not been investigated. In this paper, a simulation study is used to investigate the performance of DLMs as a measure of air pollution exposure in comparison with single- and multi-day moving average air pollution exposure measures under various forms for the true effect of air pollution on mortality. The simulation study shows that DLMs offer a more robust measure of the effect of air pollution on mortality and avoid the potential for a large negative bias compared with single- or multi-day moving average air pollution exposure measures. This is important information. In many U.S. cities, particulate air pollution concentrations are observed only once every six days, meaning it is often only possible to use single-day particulate air pollution exposure measures. The results from this paper will help quantify the magnitude of the negative bias that can result from using single-day exposure measures. The implications of this work for future air pollution mortality time series studies are discussed. The data used in this paper are concurrent daily time series of mortality, weather, and particulate air pollution from Cook County, IL, for the period 1987-1994.     

Effectiveness of cascading time series models based on meteorological factors in improving health risk prediction

Environmental Science and Pollution Research - Meteorological factors, which are periodic and regular in a long run, have an unignorable impact on human health. Accurate health risk prediction...     

A systematic literature review of deep learning neural network for time series air quality forecasting

Environmental Science and Pollution Research - Rapid progress of industrial development, urbanization and traffic has caused air quality reduction that negatively affects human health and...     

A multivariate, time series model to relate alfalfa responses to chronic, ambient sulfur dioxide exposures

A multivariate dual time series model is described here to relate chronic sulfur dioxide exposures to alfalfa responses under ambient conditions. The model considers the time series of SO(2) exposures and the growth dynamics of the crop in accounting for the final harvested biomass. The model also allows the inclusion of environmental variables such as temperature and precipitation. Coupled with the model, the 'best' regression method used facilitates the segregation or identification of the contribution of individual independent variables to the final coefficient of determination for the harvested biomass.     

Economic performance of Indonesia amidst CO2 emissions and agriculture: a time series analysis

Environmental Science and Pollution Research - To minimize the awful situation confronting the entire globe, the global warming danger has raised the intensity of consciousness from all areas of...     

PM2.5 characterization for time series studies: Pointwise uncertainty estimation and bulk speciation methods applied in Denver

Many studies have identified associations between adverse health effects and short-term exposure to particulate matter less than 2.5 μm in diameter (PM_2.5). These effects, however, are not consistent across geographical regions. This may be due in part to variations in the chemical make-up of PM_2.5 resulting from unique combinations of sources, both primary and secondary, in different regions. The Denver Aerosol Sources and Health (DASH) study is a multi-year time series study designed to characterize the daily chemical composition of PM_2.5 in Denver, identify the major contributing sources, and investigate associations between sources and a broad array of adverse health outcomes.Measurement methodology, field blank correction, pointwise uncertainty estimation and detection limit consideration are discussed in the context of bulk speciation for the DASH study. Results are presented for the first 4.5 years of mass, inorganic ion and bulk carbon speciation. The derived measurement uncertainties were propagated using the root sum of squares method and show good agreement with precision estimates derived from bi-weekly duplicate samples collected on collocated samplers. Gravimetric mass has the most uncertainty of any measurement and reconstructed mass generated from the sum of the individual species shows less uncertainty than measured mass on average. The methods discussed provide a good framework for PM_2.5 speciation measurements and are generalizable to analysis of other environmental measures.     

Forecasting of particulate matter time series using wavelet analysis and wavelet-ARMA/ARIMA model in Taiyuan,China

Particulate matter with aerodynamic diameter below 10 μm (PM₁₀) forecasting is difficult because of the uncertainties in describing the emission and meteorological fields. This paper proposed a wavelet-ARMA/ARIMA model to forecast the short-term series of the PM₁₀ concentrations. It was evaluated by experiments using a 10-year data set of daily PM₁₀ concentrations from 4 stations located in Taiyuan, China. The results indicated the following: (1) PM₁₀ concentrations of Taiyuan had a decreasing trend during 2005 to 2012 but increased in 2013. PM₁₀ concentrations had an obvious seasonal fluctuation related to coal-fired heating in winter and early spring. (2) Spatial differences among the four stations showed that the PM₁₀ concentrations in industrial and heavily trafficked areas were higher than those in residential and suburb areas. (3) Wavelet analysis revealed that the trend variation and the changes of the PM₁₀ concentration of Taiyuan were complicated. (4) The proposed wavelet-ARIMA model could be efficiently and successfully applied to the PM₁₀ forecasting field. Compared with the traditional ARMA/ARIMA methods, this wavelet-ARMA/ARIMA method could effectively reduce the forecasting error, improve the prediction accuracy, and realize multiple-time-scale prediction.

Implications: Wavelet analysis can filter noisy signals and identify the variation trend and the fluctuation of the PM₁₀ time-series data. Wavelet decomposition and reconstruction reduce the nonstationarity of the PM₁₀ time-series data, and thus improve the accuracy of the prediction. This paper proposed a wavelet-ARMA/ARIMA model to forecast the PM₁₀ time series. Compared with the traditional ARMA/ARIMA method, this wavelet-ARMA/ARIMA method could effectively reduce the forecasting error, improve the prediction accuracy, and realize multiple-time-scale prediction. The proposed model could be efficiently and successfully applied to the PM₁₀ forecasting field.     

PM2.5 characterization for time series studies: Organic molecular marker speciation methods and observations from daily measurements in Denver

Particulate matter less than 2.5 microns in diameter (PM_2.5) has been shown to have a wide range of adverse health effects and consequently is regulated in accordance with the US-EPA's National Ambient Air Quality Standards. PM_2.5 originates from multiple primary sources and is also formed through secondary processes in the atmosphere. It is plausible that some sources form PM_2.5 that is more toxic than PM_2.5 from other sources. Identifying the responsible sources could provide insight into the biological mechanisms causing the observed health effects and provide a more efficient approach to regulation. This is the goal of the Denver Aerosol Sources and Health (DASH) study, a multi-year PM_2.5 source apportionment and health study.The first step in apportioning the PM_2.5 to different sources is to determine the chemical make-up of the PM_2.5. This paper presents the methodology used during the DASH study for organic speciation of PM_2.5. Specifically, methods are covered for solvent extraction of non-polar and semi-polar organic molecular markers using gas chromatography–mass spectrometry (GC–MS). Vast reductions in detection limits were obtained through the use of a programmable temperature vaporization (PTV) inlet along with other method improvements. Results are presented for the first 1.5 years of the DASH study revealing seasonal and source-related patterns in the molecular markers and their long-term correlation structure. Preliminary analysis suggests that point sources are not a significant contributor to the organic molecular markers measured at our receptor site. Several motor vehicle emission markers help identify a gasoline/diesel split in the ambient data. Findings show both similarities and differences when compared with other cities where similar measurements and assessments have been made.     

Extreme value modeling for the analysis and prediction of time series of extreme tropospheric ozone levels: a case study

The occurrence of high concentrations of tropospheric ozone is considered as one of the most important issues of air management programs. The prediction of dangerous ozone levels for the public health and the environment, along with the assessment of air quality control programs aimed at reducing their severity, is of considerable interest to the scientific community and to policy makers. The chemical mechanisms of tropospheric ozone formation are complex, and highly variable meteorological conditions contribute additionally to difficulties in accurate study and prediction of high levels of ozone. Statistical methods offer an effective approach to understand the problem and eventually improve the ability to predict maximum levels of ozone. In this paper an extreme value model is developed to study data sets that consist of periodically collected maxima of tropospheric ozone concentrations and meteorological variables. The methods are applied to daily tropospheric ozone maxima in Guadalajara City, Mexico, for the period January 1997 to December 2006. The model adjusts the daily rate of change in ozone for concurrent impacts of seasonality and present and past meteorological conditions, which include surface temperature, wind speed, wind direction, relative humidity, and ozone. The results indicate that trend, annual effects, and key meteorological variables along with some interactions explain the variation in daily ozone maxima. Prediction performance assessments yield reasonably good results.     

Low ambient temperature increases hospital re-admissions for systemic lupus erythematosus in humid subtropical region: a time series study

Currently, the correlation between ambient temperature and systemic lupus erythematosus (SLE) hospital admissions remains not determined. The aim of this study was to explore the correlation between ambient temperature and SLE hospital admissions in Hefei City, China. An ecological study design was adopted. Daily data on SLE hospital admissions in Hefei City, from January 1, 2007, to December 31, 2017, were obtained from the two largest tertiary hospitals in Hefei, and the daily meteorological data at the same period were retrieved from China Meteorological Data Network. The generalized additive model (GAM) combined with distributed lag nonlinear model (DLNM) with Poisson link was applied to evaluate the influence of ambient temperature on SLE hospital admissions after controlling for potential confounding factors, including seasonality, relative humidity, day of week, and long-term trend. There were 1658 SLE hospital admissions from 2007 to 2017, including 370 first admissions and 1192 re-admissions (there were 96 admissions with admission status not stated). No correlation was observed between ambient temperature and SLE first admissions, but a correlation was found between low ambient temperature and SLE re-admissions (RR: 2.53, 95% CI: 1.11, 5.77) (3.5 °C vs 21 °C). The effect of ambient temperature on SLE re-admissions remained for 2 weeks but disappeared in 3 weeks. Exposure to low ambient temperature may increase hospital re-admissions for SLE, and thus it is important for SLE patients to maintain a warm living environment and avoid exposure to lower ambient temperature.