Unmasking the sky: high-resolution PM2.5 prediction in Texas using machine learning techniques.

BACKGROUND: Although PM2.5 (fine particulate matter with an aerodynamic diameter less than 2.5 µm) is an air pollutant of great concern in Texas, limited regulatory monitors pose a significant challenge for decision-making and environmental studies. OBJECTIVE: This study aimed to predict PM2.5 concentrations at a fine spatial scale on a daily basis by using novel machine learning approaches and incorporating satellite-derived Aerosol Optical Depth (AOD) and a variety of weather and land use variables. METHODS: We compiled a comprehensive dataset in Texas from 2013 to 2017, including ground-level PM2.5 concentrations from regulatory monitors; AOD values at 1-km resolution based on images retrieved from the MODIS satellite; and weather, land-use, population density, among others. We built predictive models for each year separately to estimate PM2.5 concentrations using two machine learning approaches called gradient boosted trees and random forest. We evaluated the model prediction performance using in-sample and out-of-sample validations. RESULTS: Our predictive models demonstrate excellent in-sample model performance, as indicated by high R2 values generated from the gradient boosting models (0.94-0.97) and random forest models (0.81-0.90). However, the out-of-sample R2 values fall within a range of 0.52-0.75 for gradient boosting models and 0.44-0.69 for random forest models. Model performance varies slightly across years. A generally decreasing trend in predicted PM2.5 concentrations over time is observed in Eastern Texas. IMPACT STATEMENT: We utilized machine learning approaches to predict PM2.5 levels in Texas. Both gradient boosting and random forest models perform well. Gradient boosting models perform slightly better than random forest models. Our models showed excellent in-sample prediction performance (R2 > 0.9).

Influence of distracted driving of online car-hailing drivers on overall driving performance.

The distraction affects driving performance and induces serious safety issues. To better understand distracted driving, this study examines the influence of distracted driving on overall driving performance. This paper analyzes the distraction behavior (mobile phone use, entertainment activities, and passenger interference) under three driving tasks. The statistical results show that viewing or sending messages is common during driving. Smoking, phone calls, and talking to passengers are evident in cruising, ride request and drop-off, respectively. Then, overall driving performance is proposed based on velocity, longitudinal acceleration (longacc) and yaw_rate. It is divided into three categories, high, medium, and low, by k-means algorithms. The average speed increases from low to high performance; however, the longacc and yaw_rate decrease. Finally, the influence of distracted driving on overall driving performance is analyzed using C4.5 algorithm. The result shows that when time is peak, the probability of high performance (HP) is higher than off-peak. The possibility of HP increases with the increase of duration; the number of, talking to passengers, listening to music or radio, eating; the duration of, viewing or sending messages, phone calls; but reduces with the increase of the number of phone calls. These findings provide theoretical support for driving performance evaluation.

Real-Time Driving Behavior Identification Based on Multi-Source Data Fusion.

Real-time driving behavior identification has a wide range of applications in monitoring driver states and predicting driving risks. In contrast to the traditional approaches that were mostly based on a single data source with poor identification capabilities, this paper innovatively integrates driver expression into driving behavior identification. First, 12-day online car-hailing driving data were collected in a non-intrusive manner. Then, with vehicle kinematic data and driver expression data as inputs, a stacked Long Short-Term Memory (S-LSTM) network was constructed to identify five kinds of driving behaviors, namely, lane keeping, acceleration, deceleration, turning, and lane changing. The Artificial Neural Network (ANN) and XGBoost algorithms were also employed as a comparison. Additionally, ten sliding time windows of different lengths were introduced to generate driving behavior identification samples. The results show that, using all sources of data yields better results than using the kinematic data only, with the average F1 value improved by 0.041, while the S-LSTM algorithm is better than the ANN and XGBoost algorithms. Furthermore, the optimal time window length is 3.5 s, with an average F1 of 0.877. This study provides an effective method for real-time driving behavior identification, and thereby supports the driving pattern analysis and Advanced Driving Assistance System.

Understanding and estimating the carbon dioxide emissions for urban buses at different road locations: A comparison between new-energy buses and conventional diesel buses.

Public transport buses are heavy-duty vehicles that travel through the city from morning till night, which emits a large number of greenhouse gases. Understanding and estimating the characteristics of carbon emissions for transit buses are critical in achieving a low-carbon transportation system. In this study, the changes in carbon dioxide (CO2) emissions generated from new-energy buses as well as traditional diesel buses at bus stations, intersections, and road segments are compared using statistical analysis approaches; then the factors significantly affecting the emission rates are identified based on correlation analysis and feature selection methods. Finally, a gradient boosted regression tree (GBRT) model is proposed to conduct estimations for CO2 emission rates of buses. The results indicate that different sensitivities to various influencing factors exist in the carbon dioxide emissions of different types of buses. In addition, the VT-Micro regression method and Random forest technique were utilized to compare with the developed GBRT model. According to the comparison results, the estimation errors of GBRT fluctuate in a smaller range, suggesting that the GBRT model outperforms traditional approaches in emission estimation of carbon dioxide. Also, the deep understanding of the emission characteristics for both new-energy buses and conventional diesel buses helps to plan and dispatch buses with different fuel types according to local traffic conditions.

Estimation of real-driving emissions for buses fueled with liquefied natural gas based on gradient boosted regression trees.

Nowadays, more and more conventional diesel buses are being replaced by new-energy buses in many cities in China. Although new-energy buses are more environmentally friendly compared with traditional diesel buses, they may also generate kinds of greenhouse gases as well as harmful pollutants. Currently, there exist few studies on the emission characteristics of buses with new-energy fuels, especially the liquefied natural gas (LNG) bus. The primary objective of this study is to analyze and estimate the emission rates for LNG bus in real-world driving. First, the differences in emission distribution characteristics between LNG bus and other fuel types of buses are analyzed using visualization and statistical methods. Then, a gradient boosted regression tree (GBRT) approach is applied to estimate the rates of several kinds of emissions for LNG bus, including CO, CO2, HC, and NOx, by incorporating the information of driving state in the current period and several previous periods. The performance of the developed approach is evaluated by comparing with the polynomial regression method which is widely adopted in existing literature. Experimental results demonstrate that the proposed method outperforms the competitive method for the emissions estimation of LNG bus, with the average Mean Absolute Error (MAE) reduced by 27.3%, the average Mean Absolute Percentage Error (MAPE) decreased by 33.4%, and the average Root Mean Square Error (RMSE) decreased by 22.1%. The results indicate that the proposed model is a promising approach for estimating emission rates of LNG bus. Also, this study would provide theoretical support for emission simulation tools such as MOVES, where the LNG bus emission estimation is unavailable in its current version.

Impact of freeway weaving segment design on light-duty vehicle exhaust emissions.

In the United States, 26% of greenhouse gas emissions is emitted from the transportation sector; these emisssions meanwhile are accompanied by enormous toxic emissions to humans, such as carbon monoxide (CO), nitrogen oxides (NOx), and hydrocarbon (HC), approximately 2.5% and 2.44% of a total exhaust emissions for a petrol and a diesel engine, respectively. These exhaust emissions are typically subject to vehicles' intermittent operations, such as hard acceleration and hard braking. In practice, drivers are inclined to operate intermittently while driving through a weaving segment, due to complex vehicle maneuvering for weaving. As a result, the exhaust emissions within a weaving segment ought to vary from those on a basic segment. However, existing emission models usually rely on vehicle operation information, and compute a generalized emission result, regardless of road configuration. This research proposes to explore the impacts of weaving segment configuration on vehicle emissions, identify important predictors for emission estimations, and develop a nonlinear normalized emission factor (NEF) model for weaving segments. An on-board emission test was conducted on 12 subjects on State Highway 288 in Houston, Texas. Vehicles' activity information, road conditions, and real-time exhaust emissions were collected by on-board diagnosis (OBD), a smartphone-based roughness app, and a portable emission measurement system (PEMS), respectively. Five feature selection algorithms were used to identify the important predictors for the response of NEF and the modeling algorithm. The predictive power of four algorithm-based emission models was tested by 10-fold cross-validation. Results showed that emissions are also susceptible to the type and length of a weaving segment. Bagged decision tree algorithm was chosen to develop a 50-grown-tree NEF model, which provided a validation error of 0.0051. The estimated NEFs are highly correlated with the observed NEFs in the training data set as well as in the validation data set, with the R values of 0.91 and 0.90, respectively. IMPLICATIONS: Existing emission models usually rely on vehicle operation information to compute a generalized emission result, regardless of road configuration. In practice, while driving through a weaving segment, drivers are inclined to perform erratic maneuvers, such as hard braking and hard acceleration due to the complex weaving maneuver required. As a result, the exhaust emissions within a weaving segment vary from those on a basic segment. This research proposes to involve road configuration, in terms of the type and length of a weaving segment, in constructing an emission nonlinear model, which significantly improves emission estimations at a microscopic level.

Modeling vehicle interior noise exposure dose on freeways: Considering weaving segment designs and engine operation.

Vehicle interior noise functions at the dominant frequencies of 500 Hz below and around 800 Hz, which fall into the bands that may impair hearing. Recent studies demonstrated that freeway commuters are chronically exposed to vehicle interior noise, bearing the risk of hearing impairment. The interior noise evaluation process is mostly conducted in a laboratory environment. The test results and the developed noise models may underestimate or ignore the noise effects from dynamic traffic and road conditions and configuration. However, the interior noise is highly associated with vehicle maneuvering. The vehicle maneuvering on a freeway weaving segment is more complex because of its nature of conflicting areas. This research is intended to explore the risk of the interior noise exposure on freeway weaving segments for freeway commuters and to improve the interior noise estimation by constructing a decision tree learning-based noise exposure dose (NED) model, considering weaving segment designs and engine operation. On-road driving tests were conducted on 12 subjects on State Highway 288 in Houston, Texas. On-board Diagnosis (OBD) II, a smartphone-based roughness app, and a digital sound meter were used to collect vehicle maneuvering and engine information, International Roughness Index, and interior noise levels, respectively. Eleven variables were obtainable from the driving tests, including the length and type of a weaving segment, serving as predictors. The importance of the predictors was estimated by their out-of-bag-permuted predictor delta errors. The hazardous exposure level of the interior noise on weaving segments was quantified to hazard quotient, NED, and daily noise exposure level, respectively. Results showed that the risk of hearing impairment on freeway is acceptable; the interior noise level is the most sensitive to the pavement roughness and is subject to freeway configuration and traffic conditions. The constructed NED model shows high predictive power (R = 0.93, normalized root-mean-square error [NRMSE] < 6.7%). IMPLICATIONS: Vehicle interior noise is usually ignored in the public, and its modeling and evaluation are generally conducted in a laboratory environment, regardless of the interior noise effects from dynamic traffic, road conditions, and road configuration. This study quantified the interior exposure dose on freeway weaving segments, which provides freeway commuters with a sense of interior noise exposure risk. In addition, a bagged decision tree-based interior noise exposure dose model was constructed, considering vehicle maneuvering, vehicle engine operational information, pavement roughness, and weaving segment configuration. The constructed model could significantly improve the interior noise estimation for road engineers and vehicle manufactures.

Vehicle emission implications of drivers' smart advisory system for traffic operations in work zones.

UNLABELLED: Wireless communication systems have been broadly applied in various complicated traffic operations to improve mobility and safety on roads, which may raise a concern about the implication of the new technology on vehicle emissions. This paper explores how the wireless communication systems improve drivers' driving behaviors and its contributions to the emission reduction, in terms of Operating Mode (OpMode) IDs distribution used in emission estimation. A simulated work zone with completed traffic operation was selected as a test bed. Sixty subjects were recruited for the tests, whose demographic distribution was based on the Census data in Houston, Texas. A scene of a pedestrian's crossing in the work zone was designed for the driving test. Meanwhile, a wireless communication system called Drivers Smart Advisory System (DSAS) was proposed and introduced in the driving simulation, which provided drivers with warning messages in the work zone. Two scenarios were designed for a leading vehicle as well as for a following vehicle driving through the work zone, which included a base test without any wireless communication systems, and a driving test with the trigger of the DSAS. Subjects' driving behaviors in the simulation were recorded to evaluate safety and estimate the vehicle emission using the Environmental Protection Agency (EPA) released emission model MOVES. The correlation between drivers' driving behavior and the distribution of the OpMode ID during each scenario was investigated. Results show that the DSAS was able to induce drivers to accelerate smoothly, keep longer headway distance and stop earlier for a hazardous situation in the work zone, which driving behaviors result in statistically significant reduction in vehicle emissions for almost all studied air pollutants (p-values range from 4.10E-51 to 2.18E-03). The emission reduction was achieved by the switching the distribution of the OpMode IDs from higher emission zones to lower emission zones. IMPLICATIONS: Transportation section is a significant source of greenhouse gas emissions. Many studies demonstrate that the wireless communication system dedicated for safety and mobility issues may contribute to the induction in vehicle emissions through changing driving behaviors. An insight into the correlation between the driving behaviors and the distribution of Operating Mode (OpMode) IDs is essential to enhance the emission reduction. The result of this study shows that with a Drivers Smart Advisory System (DSAS) drivers accelerated smoothly and stopped earlier for a hazardous situation, which induce the switch of the OpMode IDs from high emission zones to lower emission zones.

Impacts of pavement types on in-vehicle noise and human health.

UNLABELLED: Noise is a major source of pollution that can affect the human physiology and living environment. According to the World Health Organization (WHO), an exposure for longer than 24 hours to noise levels above 70 dB(A) may damage human hearing sensitivity, induce adverse health effects, and cause anxiety to residents nearby roadways. Pavement type with different roughness is one of the associated sources that may contribute to in-vehicle noise. Most previous studies have focused on the impact of pavement type on the surrounding acoustic environment of roadways, and given little attention to in-vehicle noise levels. This paper explores the impacts of different pavement types on in-vehicle noise levels and the associated adverse health effects. An old concrete pavement and a pavement with a thin asphalt overlay were chosen as the test beds. The in-vehicle noise caused by the asphalt and concrete pavements were measured, as well as the drivers' corresponding heart rates and reported riding comfort. Results show that the overall in-vehicle sound levels are higher than 70 dB(A) even at midnight. The newly overlaid asphalt pavement reduced in-vehicle noise at a driving speed of 96.5 km/hr by approximately 6 dB(A). Further, on the concrete pavement with higher roughness, driver heart rates were significantly higher than on the asphalt pavement. Drivers reported feeling more comfortable when driving on asphalt than on concrete pavement. Further tests on more drivers with different demographic characteristics, along highways with complicated configurations, and an examination of more factors contributing to in-vehicle noise are recommended, in addition to measuring additional physical symptoms of both drivers and passengers. IMPLICATIONS: While there have been many previous noise-related studies, few have addressed in-vehicle noise. Most studies have focused on the noise that residents have complained about, such as neighborhood traffic noise. As yet, there have been no complaints by drivers that their own in-vehicle noise is too loud. Nevertheless, it is a fact that in-vehicle noise can also result in adverse health effects if it exceeds 85 dB(A). Results of this study show that in-vehicle noise was strongly associated with pavement type and roughness; also, driver heart rate patterns presented statistically significant differences on different types of pavement with different roughness.