GPS Data and Machine Learning Tools, a Practical and Cost-Effective Combination for Estimating Light Vehicle Emissions.

This paper focuses on the emissions of the three most sold categories of light vehicles: sedans, SUVs, and pickups. The research is carried out through an innovative methodology based on GPS and machine learning in real driving conditions. For this purpose, driving data from the three best-selling vehicles in Ecuador are acquired using a data logger with GPS included, and emissions are measured using a PEMS in six RDE tests with two standardized routes for each vehicle. The data obtained on Route 1 are used to estimate the gears used during driving using the K-means algorithm and classification trees. Then, the relative importance of driving variables is estimated using random forest techniques, followed by the training of ANNs to estimate CO2, CO, NOX, and HC. The data generated on Route 2 are used to validate the obtained ANNs. These models are fed with a dataset generated from 324, 300, and 316 km of random driving for each type of vehicle. The results of the model were compared with the IVE model and an OBD-based model, showing similar results without the need to mount the PEMS on the vehicles for long test drives. The generated model is robust to different traffic conditions as a result of its training and validation using a large amount of data obtained under completely random driving conditions.

Measuring the route topography impact on real driving emissions based on neural network models.

Route topography is an important test boundary of real driving emission (RDE) tests. However, the RDE test boundaries, such as atmospheric environment, driver behavior, route topography, and traffic congestion, are random, uncertain, and completely coupled. It is difficult to know to what extent route topography can determine on-road emissions, especially in a region with hilly topography. In this regard, the neural network predictor importance algorithms were proposed to measure the importance of the route topography test boundary. Based on tens of thousands of data window samples from the RDE tests in Chongqing, factor analysis was performed to reduce the data dimensionality and eliminate information overlap, and neural network models were established to predict pollutant emissions and calculate the relative importance of input variables. The results show that route topography is comparable to trip dynamics for on-road emissions but the importance of the route topography test boundary is not fully appreciated in the existing RDE regulation, making mountain cities suffer from severe vehicle emissions that are not effectively controlled.

Estimation of Pollutant Emissions in Real Driving Conditions Based on Data from OBD and Machine Learning.

This article proposes a methodology for the estimation of emissions in real driving conditions, based on board diagnostics data and machine learning, since it has been detected that there are no models for estimating pollutants without large measurement campaigns. For this purpose, driving data are obtained by means of a data logger and emissions through a portable emissions measurement system in a real driving emissions test. The data obtained are used to train artificial neural networks that estimate emissions, having previously estimated the relative importance of variables through random forest techniques. Then, by the application of the K-means algorithm, labels are obtained to implement a classification tree and thereby determine the selected gear by the driver. These models were loaded with a data set generated covering 1218.19 km of driving. The results generated were compared to the ones obtained by applying the international vehicle emissions model and with the results of the real driving emissions test, showing evidence of similar results. The main contribution of this article is that the generated model is stronger in different traffic conditions and presents good results at the speed interval with small differences at low average driving speeds because more than half of the vehicle's trip occurs in urban areas, in completely random driving conditions. These results can be useful for the estimation of emission factors with potential application in vehicular homologation processes and the estimation of vehicular emission inventories.

Evaluation of a 10 nm Particle Number Portable Emissions Measurement System (PEMS).

On-board portable emissions measurement systems (PEMS) are part of the type approval, in-service conformity, and market surveillance aspects of the European exhaust emissions regulation. Currently, only solid particles >23 nm are counted, but Europe will introduce a lower limit of 10 nm. In this study, we evaluated a 10-nm prototype portable system comparing it with laboratory systems measuring diesel, gasoline, and CNG (compressed natural gas) vehicles with emission levels ranging from approximately 2 × 1010 to 2 × 1012 #/km. The results showed that the on-board system differed from the laboratory 10-nm system on average for the tested driving cycles by less than approximately 10% at levels below 6 × 1011 #/km and by approximately 20% for high-emitting vehicles. The observed differences were similar to those observed in the evaluation of portable >23 nm particle counting systems, despite the relatively small size of the emitted particles (with geometric mean diameters <42 nm) and the additional challenges associated with sub-23 nm measurements. The latter included the presence of semivolatile sub-23 nm particles, the elevated concentration levels during cold start, and also the formation of sub-23 nm artefacts from the elastomers that are used to connect the tailpipe to the measurement devices. The main conclusion of the study is that >10 nm on-board systems can be ready for introduction in future regulations.

Framework for the assessment of PEMS (Portable Emissions Measurement Systems) uncertainty.

European regulation 2016/427 (the first package of the so-called Real-Driving Emissions (RDE) regulation) introduced on-road testing with Portable Emissions Measurement Systems (PEMS) to complement the chassis dynamometer laboratory (Type I) test for the type approval of light-duty vehicles in the European Union since September 2017. The Not-To-Exceed (NTE) limit for a pollutant is the Type I test limit multiplied by a conformity factor that includes a margin for the additional measurement uncertainty of PEMS relative to standard laboratory equipment. The variability of measured results related to RDE trip design, vehicle operating conditions, and data evaluation remain outside of the uncertainty margin. The margins have to be reviewed annually (recital 10 of regulation 2016/646). This paper lays out the framework used for the first review of the NOx margin, which is also applicable to future margin reviews. Based on experimental data received from the stakeholders of the RDE technical working group in 2017, two NOx margin scenarios of 0.24-0.43 were calculated, accounting for different assumptions of possible zero drift behaviour of the PEMS during the tests. The reduced uncertainty margin compared to the one foreseen for 2020 (0.5) reflects the technical improvement of PEMS over the past few years.

Evaluation of NOx emissions of a retrofitted Euro 5 passenger car for the Horizon prize "Engine retrofit".

The Horizon 2020 prize for the "Engine Retrofit for Clean Air" aims at reducing the pollution in cities by spurring the development of retrofit technology for diesel engines. A Euro 5 passenger car was retrofitted with an under-floor SCR (Selective Catalytic Reduction) for NOx catalyst in combination with a solid ammonia based dosing system as the NOx reductant. The vehicle was tested both on the road and on the chassis dynamometer under various test cycles and ambient temperatures. The NOx emissions were reduced by 350-1100 mg/km (60-85%) in the laboratory depending on the test cycle and engine conditions (cold or hot start), except at type approval conditions. The reduction for cold start urban cycles was < 75 mg/km (< 15%). The on road and laboratory tests were inline. In some high speed conditions significant increase of ammonia (NH3) and nitrous oxide (N2O) were measured. No effect was seen on other pollutants (hydrocarbons, carbon monoxide and particles). The results of the present study show that retrofitting high emitting vehicles can significantly reduce vehicle NOx emissions and ultimately pollution in cities.

Exhaust particle number and composition for diesel and gasoline passenger cars under transient driving conditions: Real-world emissions down to 1.5 nm.

Recent recommendations given by WHO include systematic measurements of ambient particle number concentration and black carbon (BC) concentrations. In India and several other highly polluted areas, the air quality problems are severe and the need for air quality related information is urgent. This study focuses on particle number emissions and BC emissions of passenger cars that are technologically relevant from an Indian perspective. Particle number and BC were investigated under real-world conditions for driving cycles typical for Indian urban environments. Two mobile laboratories and advanced aerosol and trace gas instrumentation were utilized. Our study shows that passenger cars without exhaust particle filtration can emit in real-world conditions large number of particles, and especially at deceleration a significant fraction of particle number can be even in 1.5-10 nm particle sizes. The mass concentration of exhaust plume particles was dominated by BC that was emitted especially at acceleration conditions. However, exhaust particles contained also organic compounds, indicating the roles of engine oil and fuel in exhaust particle formation. In general, our study was motivated by serious Indian air quality problems, by the recognized lack of emission information related to Indian traffic, and by the recent WHO air quality guidance; our results emphasize the importance of monitoring particle number concentrations and BC also in Indian urban areas and especially in traffic environments where people can be significantly exposed to fresh exhaust emissions.

Real driving emissions data: Isuzu FTR850 AMT.

RDE is becoming a necessary element of the emissions certification of automotive vehicles. Real Driving Emissions (RDE) helps to ensure that the regular operation of a car, or heavy vehicle, is still within the acceptable emissions standards while driving under normal conditions. RDE is monitored by connecting a Portable Emissions Measurement System (PEMS) to the exhaust of the tested vehicle, which measures the pollutant concentrations as the car or truck drives along a standardised route. The data described in this paper is the raw, detailed PEMS records of a heavy goods vehicle, recorded at a rate of 1Hz, over multiple trips on an urban route in South Africa. The data includes the pollutant concentrations of CO, CO 2 , NO and NO 2 , ambient conditions, and vehicle diagnostics collected from different sensors mounted to the vehicle during the field tests. We performed no additional analysis on the data. The value of the data is in allowing researchers to (a) develop and test machine learning algorithms that predict the instantaneous pollutant concentrations or (b) studying the variance of pollutant concentrations that occurs under typical driving conditions.

Investigating the impact of route topography on real driving emission tests based on large data sample at data window level.

Route topography is an important boundary condition for the regulated real driving emission (RDE) test. However, accurately and comprehensively evaluating the influence of route topography on the RDE test is difficult, because the effect cannot be easily separated from those of other test boundaries. We selected two light-duty gasoline vehicles to complete two rounds of RDE tests on four different test routes, and conducted the correlation analysis between pollutant emissions and route topography quantified by the cumulative positive altitude gains of the test routes based on the moving averaging window method. Since the small number of sample data at the total trip and road section level were not sufficiently representative of the population, we proposed to use the pollutant emission data of the data windows to analyze the complex coupling effect of the cumulative positive altitude gains and trip dynamic parameters of v·apos[95] on the RDE tests. At data window level, thousands of data windows were treated as the road section subsets of the RDE test, and the sample space of road section emission data was expanded by several orders of magnitude. With the help of the large data sample space, the influence mechanism of the random test boundaries on the RDE tests was demonstrated.

The impacts from cold start and road grade on real-world emissions and fuel consumption of gasoline, diesel and hybrid-electric light-duty passenger vehicles.

Progressively stringent regulations regarding vehicle emissions and fuel economy have spurred technology diversification in light-duty passenger vehicles (LDPVs). To assess the real-world emissions and fuel economy performances of hybrid electric vehicles (HEVs) compared to conventional internal combustion engine (ICE) vehicles, on-road measurements of ten gasoline, four diesel and six full hybrid LDPVs were performed using portable emissions measurement systems (PEMS) in Macao, China. The hot-running emission results indicate that the high emission risks of gasoline vehicles are associated with high mileage and old model years. Diesel vehicles are found to be the highest pollutant emitters in this study due to the intentional removal of aftertreatment systems. Under hot-running conditions, HEVs, as expected, could achieve carbon-reduction benefits of approximately 30 % (i.e., lower CO2 emissions and fuel consumption) compared to their conventional gasoline counterparts, while no measurable reduction in pollutant emissions was observed except in NOX (~70 % reduction). In contrast, the cold-start extra emissions (CSEEs) of CO2 reached 120-364 g/start for these HEVs, even exceeding the maximum values of conventional gasoline vehicles. However, the higher CO2 CSEEs of HEVs can be far offset by their hot-running emission reduction benefits. For tailpipe pollutants, the CSEEs of the HEVs were reduced by 21 %-68 % on average in comparison to those of conventional gasoline vehicles. Furthermore, strong correlations (R2 values of 0.69-0.89) between the road grades and relative emissions were observed. These results can provide necessary information regarding the improvement of future LDPV emission models and inventories.

A novel method for comparing passenger car fleets and identifying high-chance gross emitting vehicles using kerbside remote sensing data.

The quantification and comparison of NOX emission from in-situ car fleets, and identification of the highest emitters is an ongoing challenge. This challenge will become more important as new and increasingly complex emissions removal systems penetrate the market. We combine real-world data with new-to-the-field statistical methods to describe fleet-scale emissions behaviours and identify candidate gross-emitter vehicles. 19,605 passenger cars were observed using a Remote Sensing Device across Aberdeen in 2015. Of these, 736 were Euro 6 Passenger Cars. The distribution of observed pollutant per unit of fuel burnt ratios for most fuel type and Euro standards followed an asymmetrical shape best characterised by the Gumbel distribution. The Gumbel distribution approach was not able to fully replicate the distribution of measurements of petrol or Euro 6 diesel cars due to the presence of a subset of high-emitting outliers, ranging from the 13th percentile for Euro 3 petrol to the 2nd percentile for Euro 6 petrol, with Euro 6 diesel having a 5th percentile outlier value. No outlier fraction was observed for pre-Euro 6 diesels. The off-model fractions resembled Gumbel distributed data and in some cases could be modelled as a separate distribution with the fleet behaving as a superposition of them. It is shown that VSP was not directly linked to this behaviour and it is hypothesised that it is caused by the emissions control systems operating sub-optimally. The reasons for sub-optimal operation are beyond the scope of this paper but may be linked to air-fuel mixture sensors, cold-start running and deterioration of the catalytic converter. Larger data-sets with more Euro 6 passenger cars are required to fully test this. Application of this methodology to larger data sets from more widely deployed remote sensing devices will allow observers to identify potentially problematic vehicles for further investigation into their emission control systems.

Impact of drivers on real-driving fuel consumption and emissions performance.

Eco-driving has attracted great attention as a cost-effective and immediate measure to reduce fuel consumption significantly. Understanding the impact of driver behaviour on real driving emissions (RDE) is of great importance for developing effective eco-driving devices and training programs. Therefore, this study was conducted to investigate the performance of different drivers using a portable emission measurement system. In total, 30 drivers, including 15 novice and 15 experienced drivers, were recruited to drive the same diesel vehicle on the same route, to minimise the effect of uncontrollable real-world factors on the performance evaluation. The results show that novice drivers are less skilled or more aggressive than experienced drivers in using the accelerator pedal, leading to higher vehicle and engine speeds. As a result, fuel consumption rates of novice drivers vary in a slightly greater range than those of experienced drivers, with a marginally higher (2%) mean fuel consumption. Regarding pollutant emissions, CO and THC emissions of all drivers are well below the standard limits, while NOx and PM emissions of some drivers significantly exceed the limits. Compared with experienced drivers, novice drivers produce 17% and 29% higher mean NOx and PM emissions, respectively. Overall, the experimental results reject the hypothesis that driver experience has significant impacts on fuel consumption performance. The real differences lie in the individual drivers, as the worst performing drivers have significantly higher fuel consumption rates than other drivers, for both novice and experienced drivers. The findings suggest that adopting eco-driving skills could deliver significant reductions in fuel consumption and emissions simultaneously for the worst performing drivers, regardless of driving experience.

Assessment of 10-nm Particle Number (PN) Portable Emissions Measurement Systems (PEMS) for Future Regulations.

The particle number (PN) emissions of vehicles equipped with particulate filters are low. However, there are technologies that can have high PN levels, especially below the currently lower regulated particle size of 23 nm. Sub-23-nm particles are also considered at least as dangerous as the larger ultrafine particles. For this reason, the European Union (EU) is planning to regulate particles down to 10 nm. In this study we compared prototype portable emission measurement systems (PEMS) and reference laboratory systems measuring from 10 nm. The tests included cycles and constant speeds, using vehicles fuelled with diesel, gasoline or liquefied petroleum gas (LPG). The results showed that the PEMS were within ±40% of the reference systems connected to the tailpipe and the dilution tunnel. Based on the positive findings and the detection efficiencies of the prototype instruments, a proposal for the technical specifications for the future regulation was drafted.

Statistical evaluation of on-road vehicle emissions measurement using a dual remote sensing technique.

On-road remote sensing (RS) is a rapid, non-intrusive and economical tool to monitor and control the emissions of in-use vehicles, and currently is gaining popularity globally. However, a majority of studies used a single RS technique, which may bias the measurements since RS only captures a snapshot of vehicle emissions. This study aimed to use a unique dual RS technique to assess the characteristics of on-road vehicle emissions. The results show that instantaneous vehicle emissions are highly dynamic under real-world driving conditions. The two emission factors measured by the dual RS technique show little correlation, even under the same driving condition. This indicates that using the single RS technique may be insufficient to accurately represent the emission level of a vehicle based on one measurement. To increase the accuracy of identifying high-emitting vehicles, using the dual RS technique is essential. Despite little correlation, the dual RS technique measures the same average emission factors as the single RS technique does when a large number of measurements are available. Statistical analysis shows that both RS systems demonstrate the same Gamma distribution with ≥200 measurements, leading to converged mean emission factors for a given vehicle group. These findings point to the need for a minimum sample size of 200 RS measurements in order to generate reliable emission factors for on-road vehicles. In summary, this study suggests that using the single or dual RS technique will depend on the purpose of applications. Both techniques have the same accuracy in calculating average emission factors when sufficient measurements are available, while the dual RS technique is more accurate in identifying high-emitters based on one measurement only.

Evaluating in-use vehicle emissions using air quality monitoring stations and on-road remote sensing systems.

This study investigated real world in-use vehicle emissions using two regulatory techniques simultaneously, namely on-road remote sensing (RS) systems and air quality (AQ) monitoring stations, aiming to provide a full pollution profile from tailpipe to roadside and atmosphere. Two large AQ and RS datasets collected during 2012-2018 were analyzed. The effects of various emission control programmes on the trends of tailpipe emissions and air quality were evaluated. Correlations between tailpipe emissions and roadside and ambient air quality were also explored. The results showed a decreasing trend of NO2 at both roadside and ambient AQ stations from 2013 to 2016, which was attributed to the intensive implementation of a series of vehicle emissions control programmes. Although NO2 was decreasing, O3 was generally increasing for all AQ stations. AQ data showed that O3 had little correlation with either NO2 or NOx, but was mainly determined by NO2/NOx ratio. Roadside NO2/NOx ratio increased first and then decreased or stabilized after 2014, while ambient NO2/NOx ratio increased steadily. RS data showed that the overall NO decreased quickly during 2012-2015 and then decreased moderately after 2015. The decrease was mainly attributed to the effective NO reduction from LPG vehicles. However, diesel NO remained high and reduced relatively slowly during the study period. Gasoline vehicles were relatively clean compared with LPG and diesel vehicles. Finally, good correlations were demonstrated between NO measured by RS sites and NOx measured by roadside AQ stations, indicating that vehicle emissions were the major contributor to roadside NOx pollution. Ambient NOx emissions could be affected by various sources, leading to different correlation levels between RS and ambient AQ results.

Fuel consumption and emissions performance under real driving: Comparison between hybrid and conventional vehicles.

Hybrid electric vehicles (HEVs) are perceived to be more energy efficient and less polluting than conventional internal combustion engine (ICE) vehicles. However, increasing evidence has shown that real-driving emissions (RDE) could be much higher than laboratory type approval limits and the advantages of HEVs over their conventional ICE counterparts under real-driving conditions have not been studied extensively. Therefore, this study was conducted to evaluate the real-driving fuel consumption and pollutant emissions performance of HEVs against their conventional ICE counterparts. Two pairs of hybrid and conventional gasoline vehicles of the same model were tested simultaneously in a novel convoy mode using two portable emission measurement systems (PEMSs), thus eliminating the effect of vehicle configurations, driving behaviour, road conditions and ambient environment on the performance comparison. The results showed that although real-driving fuel consumption for both hybrid and conventional vehicles were 44%-100% and 30%-82% higher than their laboratory results respectively, HEVs saved 23%-49% fuel relative to their conventional ICE counterparts. Pollutant emissions of all the tested vehicles were lower than the regulation limits. However, HEVs showed no reduction in HC emissions and consistently higher CO emissions compared to the conventional ICE vehicles. This could be caused by the frequent stops and restarts of the HEV engines, as well as the lowered exhaust gas temperature and reduced effectiveness of the oxidation catalyst. The findings therefore show that while achieving the fuel reduction target, hybridisation did not bring the expected benefits to urban air quality.

Solid Particle Number (SPN) Portable Emissions Measurement Systems (PEMS) in the European Legislation: A Review.

Portable emissions measurement systems (PEMS) for gaseous pollutants were firstly introduced in the United States regulation to check the in-use compliance of heavy-duty engines, avoiding the high costs of removing the engine and testing it on a dynamometer in the laboratory. In Europe, the in-service conformity of heavy-duty engines has been checked with PEMS for gaseous pollutants since 2014. To strengthen emissions regulations with a view to minimise the differences between on-road and laboratory emission levels in some cases, PEMS testing, including solid particle number (SPN), was introduced for the type-approval of light-duty vehicles in Europe in 2017 and for in-service conformity in 2019. SPN-PEMS for heavy-duty engines will be introduced in 2021. This paper gives an overview of the studies for SPN-PEMS from early 2013 with the first prototypes until the latest testing and improvements in 2019. The first prototype diffusion charger (DC) based systems had high differences from the reference laboratory systems at the first light-duty vehicles campaign. Tightening of the technical requirements and improvements from the instrument manufacturers resulted in differences of around 50%. Similar differences were found in an inter-laboratory comparison exercise with the best performing DC- and CPC- (condensation particle counter) based system. The heavy-duty evaluation phase at a single lab and later at various European laboratories revealed higher differences due to the small size of the urea generated particles and their high charge at elevated temperatures. This issue, along with robustness at low ambient temperatures, was addressed by the instrument manufacturers bringing the measurement uncertainty to the 50% levels. This measurement uncertainty needs to be considered at the on-road emission results measured with PEMS.

Characterisation of diesel vehicle emissions and determination of remote sensing cutpoints for diesel high-emitters.

Diesel vehicles are a major source of air pollutants in cities and have caused significant health risks to the public globally. This study used both on-road remote sensing and transient chassis dynamometer to characterise emissions of diesel light goods vehicles. A large sample size of 183 diesel vans were tested on a transient chassis dynamometer to evaluate the emission levels of in-service diesel vehicles and to determine a set of remote sensing cutpoints for diesel high-emitters. The results showed that 79% and 19% of the Euro 4 and Euro 5 diesel vehicles failed the transient cycle test, respectively. Most of the high-emitters failed the NO limits, while no vehicle failed the HC limits and only a few vehicles failed the CO limits. Vehicles that failed NO limits occurred in both old and new vehicles. NO/CO2 ratios of 57.30 and 22.85 ppm/% were chosen as the remote sensing cutpoints for Euro 4 and Euro 5 high-emitters, respectively. The cutpoints could capture a Euro 4 and Euro 5 high-emitter at a probability of 27% and 57% with one snapshot remote sensing measurement, while only producing 1% of false high-emitter detections. The probability of high-emitting events was generally evenly distributed over the test cycle, indicating that no particular driving condition produced a higher probability of high-emitting events. Analysis on the effect of cutpoints on real-driving diesel fleet was carried out using a three-year remote sensing program. Results showed that 36% of Euro 4 and 47% of Euro 5 remote sensing measurements would be detected as high-emitting using the proposed cutpoints.

Novel approach for connecting real driving emissions to the European vehicle laboratorial certification test procedure.

Vehicle emission certification is evaluated under laboratorial conditions, where vehicles perform a standard driving cycle in controlled conditions leading to several critics, which have resulted in the implementation of the Worldwide harmonized Light Vehicle Test Procedure (WLTP) and the Real Driving Emissions (RDE) testing procedure, as a complementary certification procedure. RDE is still under debate since boundary conditions; evaluation and trip selection methods are still being studied to allow test reproducibility. Currently, the official data analysis method uses the moving average window (MAW_EC), based on the WLTP CO2 emissions for trip validity evaluation (RDE package 4) and emissions (RDE package 3). However, this does not consider the impact of vehicle dynamics. Consequently, this work focuses on developing a novel method to relate certification driving cycle dynamics and on-road test vehicle dynamics, to evaluate RDE tests fuel use and exhaust emissions in a comparable way to certification driving cycles, indicating how close, or far, real-world driving is from the laboratorial certification test. For this, a new method was developed called road vehicle evaluation method (ROVET), which relies on the cycle vehicle dynamic and on-road trip dynamics for assessing if both tests are comparable. Results from 5 measured vehicles with a portable emissions measurement system (PEMS) through reproducibility tests and 2 case studies, show that the ROVET provides results closer to the certification calculated reference than the most commonly used method in Europe (1% avg. difference for ROVET while 8% avg. difference for MAW_EC, regarding CO2 emission, for example). The use of vehicle dynamics on construction and references of a method could be used to incentivize the regulators to review the references used by the current used methods, which suffers several criticisms since their release. As the regulated methods are in constant update, this study could be useful for helping to improve or to be used as additional method for future vehicle certification procedures. Graphical abstract.

Real-world gaseous emissions of high-mileage taxi fleets in China.

Mileage of taxi fleets is significantly higher than regular passenger vehicles in China, which might trigger greater tailpipe emissions of air pollutants. To investigate their real-world gaseous emissions, we applied portable emissions measurement systems (PEMSs) to test 44 gasoline and 24 bi-fuel taxis in seven cities. Our real-world measurement results indicated that a major part of the tested China 3 and China 4 gasoline taxis, especially the samples with high mileage (>300,000 km), far exceeded the corresponding emission limits of NOX, THC and CO. Only the newest China 5 gasoline taxis with relatively lower mileage had effective emission controls and the gaseous emissions were below the limits. Illegal tampering, malfunction and deterioration of three-way catalytic converters (TWC) are major reasons for high emissions from high-mileage taxis. First, China 4 gasoline taxis without TWC (purposely removed by drivers) increased their gaseous emissions than TWC-equipped counterparts by more than one order of magnitude. Second, bi-fuel taxis when using compress natural gas (CNG) had much higher NOX and THC emissions than those when using gasoline, which might be probably attributed to unsophisticated engine calibration and unfavorable TWC working conditions. Furthermore, TWC renewal could bring immediate and substantial emission reductions (up to 70%) for high-mileage taxis. However, such benefits from TWC renewal would become less significant as the mileage levels further increase. We also found a good correlation between CO and THC emissions for gasoline taxis, whose cold start effects were both significant. This study poses significant concerns regarding real-world emissions of high-mileage taxi fleets in China, which could consist of many gross emitters in the urban areas. Stringent in-use compliance programs and in particular frequent TWC renewals for high-mileage taxis should be implemented by policy makers in China.