Effect of Altitude on Gasoline Combustion Efficiency in Light-Duty Vehicles: Evidence from Andean Corridors (0–4000 m a.s.l.)

Abstract

Reduced air density at altitude challenges spark-ignition combustion in light-duty vehicles, particularly in Andean countries where fleets operate between sea level and over 4000 m a.s.l. We quantified the effect of altitude on gasoline combustion efficiency and pollutant formation using 94,263 second-by-second records (2021–2025) from ten Euro 2–5 vehicles travelling Andean and coastal corridors in Ecuador (0–4000 m a.s.l.). Emissions were measured with a Brain Bee AGS-688 analyser with pressure compensation; fuel consumption via OBD-II; operational demand via smoothed Vehicle Specific Power (${\mathrm{PSV}}_{\mathrm{m}10}$) in six K-Means clusters. We propose $R_{\mathrm{CO}}=\left[\mathrm{CO}\right]/\left[{\mathrm{CO}}_2\right]$ and $R_{\mathrm{HC}}=\left[\mathrm{HC}\right]/\left[{\mathrm{CO}}_2\right]$ as standardised, displacement-independent combustion quality indicators. $R_{\mathrm{CO}}$ at 3500–4000 m exceeded the 500–1000 m value by 7.5×, and $R_{\mathrm{HC}}$ by 18×. Nitric oxide showed a non-linear “N”-pattern, peaking locally at 2000–2500 m (212 ppm) and absolutely at 3500–4000 m (613 ppm), linked to EGR suppression and rising chamber temperature. CO emission factors reached 6.42 g/km at 2000–2500 m versus 1.32 g/km at the coast (4.9×). These results provide the first naturalistic evidence base for calibrating high-altitude emission inventories in Andean corridors.

1. Introduction

Road transport is one of the most relevant sources of urban air pollution worldwide, with light-duty gasoline vehicles being the primary emitters of carbon monoxide (CO), unburned hydrocarbons (HC) and nitrogen oxides (${\mathrm{NO}}_x$) in most developing economies [1]. In Latin America, sustained fleet growth has intensified pressure on air quality in cities combining high vehicle density with severe topographic conditions [2]. In Ecuador, the light vehicle fleet exceeds three million units, of which over 60% are spark-ignition engines with fuel management calibrations lacking adaptive altitude compensation [3].

A first-order physical complication arises in countries with pronounced topographic relief: as altitude increases, atmospheric pressure decreases, reducing intake air density and the oxygen mass available for combustion per cycle [4,5]. In engines without adaptive altitude management—common in Euro 2 and Euro 3 calibrations—the shift of the air–fuel ratio towards richer mixtures produces incomplete oxidation of carbon and hydrogen, raising CO and HC in the exhaust [6,7,8]. This effect was documented in 1986 in Denver, Colorado (≈1600 m a.s.l.), where [9] showed that CO was the most altitude-sensitive pollutant fraction. Ref. [7] reported significant increases in CO and VOC between 0 and 3500 m in India; [10] measured, using PEMS, that CO and ${\mathrm{CO}}_2$ emission factors grow between 2270 and 4540 m on the Tibetan Plateau; and [11] quantified a CO increase of 2.56 times from 700 to 2400 m in China VI vehicles under RDE conditions. These studies, however, differ markedly in scope and design—bench dynamometer tests [6], short altitude transects (0–3500 m) [7], and single-region RDE campaigns [11]—so that the shape of the CO–altitude relationship over a single continuous 4000 m gradient, and its dependence on Euro emission standard, remains empirically unresolved.

The behaviour of nitric oxide (NO) with altitude is more complex. The reduction in peak cylinder temperature and the progressive suppression of exhaust gas recirculation (EGR) modify the Zeldovich thermal pathway [4]. Ref. [8] documented, using the MEDAS atmosphere simulator between 150 and 3000 m, that ${\mathrm{NO}}_x$ grows up to 2000 m—where the EGR valve closes completely—and then stabilises or decreases. Ref. [11] report an “N”-shaped distribution with a peak at ≈1900 m, consistent with the present study’s findings. Ref. [12] found that road ${\mathrm{NO}}_x$ can triple urban values in light-duty diesel vehicles at altitude.

Regarding fuel consumption, [13] systematically studied altitude–consumption interaction in a Euro 3 gasoline vehicle using an altimetric test bench, finding a non-monotonic effect: −3.5% in NEDC and +6.2% in Highway at 2200 m. Ref. [14] extended the analysis to 3030 m using a GT-Power model, showing that above 80 km/h and 2800 m engine performance deterioration outweighs aerodynamic resistance reduction. Ref. [15] confirm this behaviour via OBD-II data in naturalistic driving in the Ecuadorian Andean corridor. Because [13,14] both rely on controlled bench or simulated conditions rather than real-world driving, whether their non-monotonic consumption response holds under naturalistic operational-demand variability—where acceleration events and grade are not prescribed—has not been directly tested over a continuous altitude gradient.

The Real Driving Emissions (RDE) methodology, enabled by Portable Emission Measurement Systems (PEMS), has advanced the understanding of pollutant formation under real operating conditions [16,17,18]. However, most RDE studies have been conducted at altitudes below 1500 m [19,20], with only a small group on the Tibetan Plateau [10]. In Latin America, existing high-altitude studies are limited to diesel buses [18] or static measurements at a single altitude [3,21]. Studies covering 0–4000 m in a single campaign with a naturalistic protocol, mixed Euro 2–5 fleet, and second-by-second recording are non-existent in the literature.

Vehicle Specific Power (VSP) has proven superior to approaches based solely on speed and acceleration for estimating real-world emission rates [17,22,23]. Its smoothed variant ${\mathrm{PSV}}_{\mathrm{m}10}$ (10 s moving median) reduces sensitivity to GPS noise peaks [15,24]. Unlike the VSP bins of the MOVES model—whose inadequacy for mountainous terrain was demonstrated by [17]—K-Means clustering on ${\mathrm{PSV}}_{\mathrm{m}10}$ yields empirically representative clusters of Andean conditions.

The ratios CO/${\mathrm{CO}}_2$ and HC/${\mathrm{CO}}_2$ have been proposed as robust combustion efficiency indicators normalised to fuel consumed [25], independent of engine displacement and dilution effects. Ref. [18] proposed fuel-based emission factors (${\mathrm{EF}}_{\mathrm{i}}^{*}$, g/kg) with analogous logic for diesel buses, reducing inter-region variability by a factor of three. The systematic extension to CO/${\mathrm{CO}}_2$ and HC/${\mathrm{CO}}_2$ stratified by altitudinal band has not been reported in the literature.

Taken together, three specific gaps limit the current evidence base. First, no naturalistic real-driving study has covered a continuous 0–4000 m gradient in a single measurement campaign: existing RDE/PEMS work is concentrated below 1500 m [19,20], the only higher-altitude campaign is confined to the Tibetan Plateau’s already-high baseline (2270–4540 m, without a sea-level reference) [10], and Andean evidence is based on static, single-altitude testing rather than continuous driving [3,21]—precluding within-study comparison across the full range over which atmospheric pressure more than halves. Second, Latin American high-altitude evidence for spark-ignition vehicles specifically is essentially absent: the region’s only naturalistic high-altitude RDE study to date targeted diesel buses [18], whose combustion regime, aftertreatment and altitude sensitivity differ fundamentally from light-duty gasoline engines. Third, existing altitude–emission studies have not simultaneously controlled for operational demand: ${\mathrm{PSV}}_{\mathrm{m}10}$-based stratification, shown elsewhere to be essential for isolating the altitude effect from confounding driving-style variance [17], has not previously been combined with a full-gradient altitude design, leaving open the possibility that reported altitude effects partly reflect uncontrolled differences in driving behaviour between low- and high-altitude test sites. This study addresses all three gaps directly, with a single naturalistic campaign spanning 0–4000 m a.s.l. in gasoline light-duty vehicles specifically, under ${\mathrm{PSV}}_{\mathrm{m}10}$-based operational-demand control throughout.

This study analyses 94,263 second-by-second records (2021–2025) from ten light-duty Euro 2–5 vehicles in Ecuadorian Andean corridors between 0–4000 m a.s.l. The specific objectives are:

• To quantify the effect of altitude on CO, ${\mathrm{CO}}_2$, HC and NO along a continuous 4000 m gradient under real driving conditions.
• To characterise how operational demand (${\mathrm{PSV}}_{\mathrm{m}10}$, 6 K-Means clusters) modulates the altitude–emission relationship.
• To evaluate fuel consumption as a function of altitude and operational demand using OBD-II data.
• To propose and validate $R_{\mathrm{CO}}$ and $R_{\mathrm{HC}}$ as standardised combustion quality indicators by altitudinal band for mixed Euro-standard fleets.

2. Materials and Methods

2.1. Study Area and Road Corridors

Data were collected along five interconnected road corridors in Ecuador spanning from sea level to ≈4000 m a.s.l. (Figure 1). The corridors cross the Pacific coastal plain, the inter-Andean valley and the high-altitude paramo, including urban, peri-urban and rural segments. Within less than 200 km, atmospheric pressure drops from 101.3 to 61.6 kPa [5], a gradient comparable to the Tibetan Plateau [10] but additionally covering low altitudes (0–800 m) absent in all previous Latin American Andean studies. The five corridors were: Riobamba–Guano, Riobamba–Chimborazo, Riobamba–Santo Domingo, Riobamba–Guayaquil and Santo Domingo–El Carmen. These routes combine two-lane paved rural highways (predominant surface throughout), urban arterials with signalised intersections in Riobamba, Guayaquil and Santo Domingo, and peri-urban segments with mixed heavy- and light-duty traffic; posted speed limits range from 30 km/h in urban segments to 90 km/h on inter-urban highway sections. Recording sessions were distributed across all times of day and both dry and wet-season conditions during 2021–2025, so that the dataset reflects the range of traffic density and ambient conditions the instrumented fleet encounters in normal operation, rather than a single controlled test day.

2.2. Test Vehicle Fleet

Ten light-duty gasoline vehicles were instrumented according to the criteria: (i) displacement 1000–2500 cm³; (ii) Euro 2–5 emission standard; (iii) OBD-II compliance; (iv) regular operational use on the study corridors during 2021–2025. The deliberate inclusion of Euro 2–5 vehicles—unlike studies with homogeneous fleets [11,17]—allows evaluation of the moderating effect of emission control technology on the altitudinal response (

Table 1. Main characteristics of the test vehicle fleet.

ID	Fuel	Displacement (${\mathrm{cm}}^3$)	Euro	Fuel mgmt.	Year	n (Records)
V01	Gasoline	1400	2	MPFI	2010	8742
V02	Gasoline	1600	3	MPFI	2012	11,318
V03	Gasoline	1800	3	MPFI	2013	9215
V04	Gasoline	1600	4	MPFI	2015	10,481
V05	Gasoline	2000	4	MPFI	2016	9876
V06	Gasoline	1000	4	MPFI	2017	8954
V07	Gasoline	2000	5	MPFI	2019	11,632
V08	Gasoline	2500	5	MPFI	2020	10,287
V09	Gasoline	1800	5	MPFI	2022	9143
V10	Gasoline	1600	5	MPFI	2023	10,385
Total						100,033

MPFI: Multi-point fuel injection.

). Vehicles were recruited from private owners and institutional fleets operating regularly on the study corridors, so that instrumentation captured genuine in-service use rather than dedicated test drives; the Brain Bee AGS-688 and OBD-II logger (Section 2.3.1) were installed for the duration of each vehicle’s normal daily circulation, and data were retained only for trips consistent with the vehicle’s routine usage pattern. Table 1 reports fuel management technology and model year as proxies for emission-control sophistication; kilometrage, detailed technical-inspection status and after-treatment verification per vehicle were not part of the original data-collection protocol and are acknowledged as a limitation (Section 4.5).

2.3. Measurement Equipment and Data Acquisition

2.3.1. Exhaust Gas Analyser

Exhaust gas concentrations were measured with a Brain Bee AGS-688 analyser (Brain Bee S.p.A., Parma, Italy) operating via NDIR for CO, ${\mathrm{CO}}_2$ and HC (n-hexane equivalent), and electrochemical sensor for NO. The instrument complies with ISO 3930/OIML Class 0 [26,27] and Directive 2004/22/EC [28]. It incorporates automatic ambient pressure compensation between 85.0 and 106.0 kPa (manufacturer specification [29]), corresponding to elevations from sea level up to approximately 1500 m a.s.l. under ISA/ICAO Doc 7488 conditions [30]—unlike conventional PEMS used in similar studies [17,18] that require external barometric corrections. Data were recorded at 1 Hz via RS-232 (

Table 2. Technical specifications of the Brain Bee AGS-688 analyser.

Channel	Range	Resolution	Uncertainty (MPE)	Response Time	Principle
CO	0–9.99 % vol	0.01 % vol	±0.03 % vol or ±5% b	<10 s	NDIR
${\mathrm{CO}}_2$	0–19.9 % vol	0.10 % vol	±0.5 % vol or ±5% b	<10 s	NDIR
HC	0–9999 ppmvol	1 ppmvol	±10 ppmvol or ±5% b	<10 s	NDIR
NO	0–5000 ppmvol	1 ppmvol	n/a c,d	<60 s	Electrochemical

NDIR: Non-dispersive infrared. Pressure compensation: 85.0–106.0 kPa. ^b Maximum permissible error (MPE) at initial verification for OIML R99 Class 0 instruments (absolute or relative value, whichever is greater) [26]; the Brain Bee AGS-688 is certified to this class (Section 2.3.1). ^c OIML R99 does not define a certified accuracy class for the NO channel (it is used in R99 only as an interference-test gas for the classed channels); no independent manufacturer accuracy specification for the electrochemical NO cell was available in public documentation, which is stated explicitly as a residual limitation (Section 4.5). ^d In a companion measurement campaign using the same instrument class over the same Andean corridors [31], field cross-validation of the electrochemical NO cell against a reference PEMS (parSYNC^® FLEX-PNC) found an average difference of ±30–50 ppm at altitudes above 2500 m a.s.l., within the admissible range for portable electrochemical sensors under ISO 16183 [32]. This cross-validation was not repeated for the specific instrument units used in the present study, so it is reported here as supporting evidence rather than a direct calibration record for this dataset.

).

Above ≈1500 m a.s.l., barometric pressure in the present dataset fell to as low as 54.9 kPa (Section 3), i.e., below the analyser’s certified compensation range. Raw volumetric concentration readings (%vol, ppmvol) recorded above this altitude may therefore carry a systematic uncertainty that is not resolved by the instrument’s internal compensation. No certified span-gas verification was performed at high altitude in the field; this limitation is stated explicitly in Section 4.5 rather than corrected post hoc. Independently of the analyser’s internal pressure compensation, the exhaust mass flow used to convert volumetric concentrations into mass emission rates was corrected for local air density when derived via the instantaneous-fuel-consumption method (Section 2, Equation (5)). This correction addresses only the mass-flow calculation. It is not equivalent to, nor a substitute for, the analyser’s own compensation of the raw concentration signal.

2.3.2. OBD-II Fuel Consumption and Vehicle Kinematics

Instantaneous fuel consumption was acquired via OBD-II PID 0x5E (Engine Fuel Rate, L/h). This methodology showed ${\mathrm{R}}^2>0.90$ correlation with gravimetric measurements in high-altitude buses [18]. The signal was synchronised with the gas analyser to ±0.5 s. Vehicle speed was obtained from OBD-II (PID 0x0D) and GPS; instantaneous acceleration was calculated as a first-order finite difference filtered with a 3 s moving average [17].

2.4. Emission Factor Calculation

Mass emission rates (g/s) were derived from volumetric gas concentrations, instantaneous fuel rate and vehicle speed using two complementary routes, selected automatically on a record-by-record basis: an instantaneous-fuel-consumption (IFC) route, used for the large majority of records, and a direct air-mass-flow (MAF) route, used when the IFC route is not defined (Section 2.4.5). The full derivation and all numerical assumptions are given below so that the calculation can be independently reproduced.

2.4.1. Exhaust Mass Flow Rate (IFC Route)

Instantaneous fuel mass flow was obtained from the OBD-II fuel rate ${\dot{V}}_f$ (L/h, PID 0x5E) and fuel density ${\rho }_f$:

$${\dot{m}}_f={\displaystyle \frac{{\dot{V}}_f}{3600}}{\rho }_f,$$

(1)

with ${\rho }_f=720$ g/L, representative of commercial gasoline at ambient conditions [4]. Total (wet) exhaust mass flow was estimated from a stoichiometric mass balance using the real-time, not nominal, air–fuel ratio:

$${\dot{m}}_{\mathrm{exh},\mathrm{wet}}=\left(1+{\mathrm{AFR}}_{\mathrm{real}}\right){\dot{m}}_f,{\mathrm{AFR}}_{\mathrm{real}}=\lambda ·{\mathrm{AFR}}_{\mathrm{stoich}},$$

(2)

where $\lambda$ is the instantaneous relative air–fuel ratio (Section 3.3) and ${\mathrm{AFR}}_{\mathrm{stoich}}=14.7$ for gasoline. Using the real-time $\lambda$ rather than the nominal stoichiometric constant is essential in this dataset, since $\lambda$ departs from unity systematically (median 1.14–1.25 across altitudinal bands); using a fixed stoichiometric AFR would bias the exhaust flow estimate by the same margin.

2.4.2. Wet-to-Dry Basis Correction

The Brain Bee AGS-688 measures gas concentrations on a dry basis: water vapour is removed by the condensate separator upstream of the sensors (Section 2.3.1). Equation (2), by contrast, yields the wet total exhaust mass flow, since it balances all combustion products including ${\mathrm{H}}_2\mathrm{O}$. Volumetric concentrations (dry basis) must therefore be combined with a dry-basis exhaust mass flow. Assuming complete combustion of a representative fuel/unburned-hydrocarbon surrogate ${\mathrm{CH}}_y$ ($y=1.85$, consistent with the hydrocarbon mass convention of Section 2.4.4) at relative air–fuel ratio $\lambda$:

$${\mathrm{CH}}_y+\lambda \left(1+\frac{y}{4}\right)\left({\mathrm{O}}_2+3.76{\mathrm{N}}_2\right)\to {\mathrm{CO}}_2+\frac{y}{2}{\mathrm{H}}_2\mathrm{O}+(\lambda -1)\left(1+\frac{y}{4}\right){\mathrm{O}}_2+3.76\lambda \left(1+\frac{y}{4}\right){\mathrm{N}}_2,$$

(3)

the mass fraction of water in the wet exhaust, $w\left(\lambda \right)$, follows directly from the molar amounts and molecular weights of Equation (3), and the dry exhaust mass flow used in Equation (6) is:

$${\dot{m}}_{\mathrm{exh},\mathrm{dry}}={\dot{m}}_{\mathrm{exh},\mathrm{wet}}\left[1-w\left(\lambda \right)\right].$$

(4)

For the $\lambda$ range observed in this study, $w\left(\lambda \right)\approx 0.06$–$0.07$ (6–7% of wet exhaust mass), consistent with typical values for stoichiometric-to-lean gasoline combustion [4].

2.4.3. Altitudinal Density Correction

For IFC-route records, the dry exhaust mass flow was further scaled by the local-to-standard air density ratio (Section 2.3.1):

$$f_c={\displaystyle \frac{{\rho }_{\mathrm{local}}}{{\rho }_0}},{\rho }_{\mathrm{local}}={\displaystyle \frac{P_{\mathrm{local}}}{R_a{T}_{\mathrm{intake}}}},$$

(5)

where $P_{\mathrm{local}}$ is local barometric pressure (ISA/ICAO Doc 7488 [30]), $T_{\mathrm{intake}}$ is intake air temperature (OBD-II PID 0x0F, K), $R_a=287.05$ J ${\mathrm{kg}}^{-1}{\mathrm{K}}^{-1}$ is the specific gas constant of air, and ${\rho }_0=1.225$ kg ${\mathrm{m}}^{-3}$ is the ISA sea-level reference density. MAF-route records were not scaled by $f_c$, since the hot-wire air-mass-flow sensor measures actual local air mass directly and already reflects ambient density.

2.4.4. Species Mass Emission Rate

The mass emission rate of species i is:

$${\dot{m}}_i={\dot{m}}_{\mathrm{exh}}·x_i·{\displaystyle \frac{M_i}{\overline{M}}},$$

(6)

where ${\dot{m}}_{\mathrm{exh}}={\dot{m}}_{\mathrm{exh},\mathrm{dry}}·f_c$ for IFC-route records (Equations (4) and (5)), or the directly measured air mass flow plus ${\dot{m}}_f$ for MAF-route records; $x_i$ is the dry-basis mole fraction of species i (concentration/100 for % vol channels CO, ${\mathrm{CO}}_2$, ${\mathrm{O}}_2$; concentration/${10}^6$ for ppmvol channels HC, NO); $M_i$ is the molecular weight of species i; and $\overline{M}=28.96$ g/mol is the mean molecular weight of dry exhaust, approximated by that of dry air [4].

Table 3. Molecular weights used in Equation (6).

Species	${\mathit{M}}_{\mathit{i}}$ (g/mol)	Basis
CO	28.00	Molecular weight of CO
${\mathrm{CO}}_2$	44.00	Molecular weight of ${\mathrm{CO}}_2$
HC	13.876	${\mathrm{CH}}_{1.85}$ equivalent (regulatory THC mass convention)
NO	30.00	Molecular weight of NO (not ${\mathrm{NO}}_2$-equivalent)

lists the values of $M_i$ used.

Two conventions in Table 3 require explicit justification. First, although the analyser’s NDIR channel is calibrated against n-hexane (Section 2.3.1), the ppmvol readout is converted to mass using the standard regulatory hydrocarbon surrogate ${\mathrm{CH}}_{1.85}$ ($M=13.876$ g/mol), not the molecular weight of hexane itself (86.18 g/mol); the calibration gas fixes the NDIR response curve, while the reporting convention fixes the mass basis, and the two are independent. Second, only NO is reported (electrochemical channel), not total ${\mathrm{NO}}_{\mathrm{x}}$ as ${\mathrm{NO}}_2$-equivalent (46 g/mol); this choice reflects the sensor’s actual measurand and is used consistently throughout the manuscript (see Abbreviations in the back matter).

2.4.5. Idle Records and Method Selection

The IFC route is undefined when vehicle speed is zero, since the logged fuel-economy field (L/100 km) is computed downstream as a ratio to distance and is not reported at $v=0$. In the present dataset, 13,460 records (13.1%) were logged at $v=0$ km/h; for all of these, the pipeline automatically falls back to the MAF route, which uses the directly measured intake air mass flow and is independent of vehicle speed. Only 288 idle records (2.1% of idle records; <0.3% of the full dataset) yielded a null mass-flow estimate under both routes and were excluded from emission-factor aggregation as a data-quality artefact (Section 2.6).

2.4.6. From Instantaneous Mass Rate to Distance-Based Emission Factor

Instantaneous emission factors (g/km) were computed as ${\mathrm{EF}}_i={\dot{m}}_i/(v/3600)$ for records with $v>0$, and aggregated as the band-wise median reported in Section 3. Records at $v=0$ contribute to the cumulative mass balance of a trip but, having no associated distance, are excluded from the instantaneous ${\mathrm{EF}}_i$ distribution, consistent with standard practice [18].

2.4.7. Validation of the Calculation Chain

Because Equations (2) and (4) refine earlier internal calculations that used a nominal (rather than real-time) air–fuel ratio and did not apply an explicit wet-to-dry correction, we verified their impact by recomputing ${\dot{m}}_i$ for all IFC-route records under both formulations. The refined values differ from those used throughout Results by a modest and statistically stable margin (median $-10\%$ to $-6\%$ across altitudinal bands and species, CO, ${\mathrm{CO}}_2$, HC and NO alike), without altitude-dependent trend. This confirms that the altitudinal patterns reported in Section 3 are not an artefact of the nominal-AFR/wet-basis simplification, and bounds the associated uncertainty in absolute EF magnitudes to approximately ±10% (Section 4.5).

2.5. Data Collection Protocol

Measurements were conducted under naturalistic driving conditions without prescribed cycle or speed restriction [16,18], distributed throughout the 2021–2025 campaign. All sessions were performed with the engine at normal operating temperature (coolant ≥ 80 °C) [17]. Each session logged the Brain Bee AGS-688 exhaust channels and the OBD-II/GPS kinematic channels concurrently and continuously, from key-on to key-off. Acquisition rates are given in Section 2.3.1 (1 Hz gas analyser, OBD-II polling synchronised to the same 1 Hz base rate). No separate trip segmentation or manual annotation was required: every logged second of engine operation is represented by one record in the raw dataset, prior to the quality-control filtering described in Section 2.6. Consistent with the naturalistic-driving design, sessions covered the full range of routine driving purposes for each vehicle (commuting, intercity travel and incidental urban trips), rather than a fixed number of prescribed repetitions per corridor.

2.6. Data Quality Control

A total of 94,263 valid records in the 0–4000 m range were obtained after the following filters, applied sequentially to the full logged dataset (103,033 raw records logged in total, including sessions extending above the 4000 m a.s.l. study boundary; 8770 records, 8.5%, fell above 4100 m a.s.l. and were excluded by filter (3) alone (4000–4500 m: 4804 records with normal moving-vehicle data; 4500–5000 m: 3966 records, entirely stationary/idle at the single highest-altitude stop on the corridor), both outside the nominal 4000 m study boundary. Filters (1), (2), (4) and (5) were applied upstream during data consolidation; the number of records each individually removed was not separately logged and cannot be reconstructed retroactively, which is acknowledged as a limitation of the data-processing record):

(1)

Exclusion of cold-start (coolant < 80 °C). Cold-start enrichment produces transiently elevated CO and HC unrelated to steady-state altitude effects and would otherwise confound the altitudinal comparison.

(2)

Removal of saturations: CO > 9.5 % vol or HC > 9500 ppmvol. These thresholds correspond to the upper 95% of the analyser’s certified range (Table 2), beyond which linearity is not guaranteed.

(3)

Exclusion of GPS altitude outside 0–4100 m a.s.l. A 100 m margin above the nominal 4000 m upper bound was allowed to avoid discarding valid records due to GPS vertical-positioning noise near the ceiling of the study range.

(4)

Removal of duplicate timestamps and implausible acceleration ($\left|a\right|>4$ m ${\mathrm{s}}^{-2}$). The acceleration bound exceeds the practical performance envelope of the instrumented light-duty fleet and flags GPS/OBD-II synchronisation artefacts (Section 2.3.1) rather than genuine vehicle dynamics.

(5)

Discard of records with ${\mathrm{CO}}_2$ < 2 % vol. Values below this threshold are inconsistent with a firing engine under load and indicate probe disconnection or a sampling-line leak.

2.7. Vehicle Specific Power and Operational Clustering

2.7.1. Definition of ${\mathrm{PSV}}_{\mathrm{m}10}$

VSP is defined as [22]:

$$\mathrm{VSP}=v\left[a(1+{\epsilon }_i)+g\sin \theta +g{C}_R\cos \theta \right]+{\displaystyle \frac{1}{2}}{\rho }_a{C}_D{A}_f{\displaystyle \frac{v^3}{m}}$$

(7)

where v (m/s) is speed, a (${\mathrm{m/s}}^2$) acceleration, ${\epsilon }_i=0.10$, g = 9.81 m/s², $\theta$ (rad) road grade, $C_R=0.015$, ${\rho }_a$ air density, $C_D$ aerodynamic coefficient, $A_f$ (${\mathrm{m}}^2$) frontal area and m (kg) kerb mass. The 10 s moving median:

$${\mathrm{PSV}}_{\mathrm{m}10}\left(t\right)=\mathrm{median}\left\{\mathrm{VSP}(t-9),\dots ,\mathrm{VSP}\left(t\right)\right\}$$

(8)

confers robustness against GPS noise peaks [24]. Air density ${\rho }_a$ was estimated second-by-second using the ISA model [5]:

$${\rho }_a\left(h\right)={\rho }_0{\left(1-{\displaystyle \frac{Lh}{T_0}}\right)}^{gM/\left(RL\right)-1}$$

(9)

with ${\rho }_0=1.225 {\mathrm{kg}/\mathrm{m}}^3$, $T_0=288.15$ K, $L=0.0065$ K/m.

2.7.2. K-Means Clustering

The dataset was segmented into six clusters using K-Means on ${\mathrm{PSV}}_{\mathrm{m}10}$ and instantaneous acceleration, both standardised. The optimal $k=6$ was determined with the elbow criterion and silhouette coefficient. Clusters were labelled: Idle/stopped, Deceleration, Light cruise, Positive acceleration, High demand and Maximum demand, following the taxonomy of [23]. Standardisation (zero mean, unit variance) was applied to both features prior to clustering so that ${\mathrm{PSV}}_{\mathrm{m}10}$—expressed in W/kg and spanning a substantially wider numerical range than acceleration in ${\mathrm{m/s}}^2$—would not dominate the Euclidean distance metric used by K-Means. The two-feature design deliberately mirrors the operational variables that determine tractive power demand (Equation (7)), without embedding altitude or gas concentration in the clustering step. As a result, the resulting operational-mode clusters can be used as an independent stratifying variable when subsequently testing for an altitude effect. This avoids circularity between the exposure of interest and the covariate used to control for driving style.

2.7.3. Road Gradient

Road grade $\theta$ enters the ${\mathrm{PSV}}_{\mathrm{m}10}$ formulation directly (Equation (7), $g\sin \theta$ term) but its independent association with emissions had not been previously stratified. Grade was computed second-by-second from GPS-derived elevation and distance increments and expressed both as an angle (rad) and as a percentage slope ($100\tan \theta$). Because the underlying elevation/distance ratio is ill-conditioned at near-zero horizontal displacement, a subset of records showed non-physical values, up to and including the $\tan \theta$ singularity at $\theta =\pm \pi /2$; these are GPS artefacts (intermittent multipath/shadowing in mountainous terrain), not real road geometry. Records with $\left|\mathrm{slope}\right|>30\%$—a generous bound relative to the steepest sustained grades on paved Andean roads [3]—were treated as invalid and excluded from gradient-specific analyses only (84,853 of 94,263 records, 90.0%, remained valid). For stratified comparison, valid records were classified as downhill ($\mathrm{slope}\le -3\%$), flat ($-3\%<\mathrm{slope}<3\%$) or uphill ($\mathrm{slope}\ge 3\%$).

2.8. Combustion Quality Indicators

$$R_{\mathrm{CO}}={\displaystyle \frac{\left[\mathrm{CO}\right]}{\left[{\mathrm{CO}}_2\right]}},R_{\mathrm{HC}}={\displaystyle \frac{\left[\mathrm{HC}\right]}{\left[{\mathrm{CO}}_2\right]}}$$

(10)

Concentrations in consistent % vol (ppmvol/${10}^4$ for HC). Analogous to the ${\mathrm{EF}}_{\mathrm{i}}^{*}$ (g/kg fuel) factors of [18]; do not require exhaust mass flow measurement [25]. Because ${\mathrm{CO}}_2$ tracks the carbon mass actually oxidised to completion, normalising CO and HC to ${\mathrm{CO}}_2$ rather than to intake air or total exhaust volume expresses each ratio as unburned/partially-oxidised carbon relative to burned carbon, independent of engine displacement, dilution by excess air, and—by construction—of the exhaust volumetric flow rate, which is not directly measured by an exhaust-pipe-probe analyser such as the Brain Bee AGS-688 (Section 2.3.1). $R_{\mathrm{CO}}$ and $R_{\mathrm{HC}}$ are therefore well-suited to a mixed Euro-standard fleet, where displacement and injection technology otherwise complicate comparison of raw concentrations or absolute mass emission rates across vehicles. Both ratios are dimensionless and were computed from the same synchronised 1 Hz records used throughout (Section 2.3.1), so no additional temporal alignment was required.

2.9. Altitudinal Band Stratification and Statistical Analysis

The altitudinal range was divided into eight 500 m bands (

Table 4. Altitudinal bands with ISA atmospheric pressure and air density.

Band	Altitude (m a.s.l.)	Pressure (kPa)	${\mathit{\rho }}_{\mathit{a}}$ (${\mathrm{kg m}}^{-3}$)
B1	0–500	101.3–95.5	1.225–1.167
B2	500–1000	95.5–89.9	1.167–1.112
B3	1000–1500	89.9–84.6	1.112–1.058
B4	1500–2000	84.6–79.5	1.058–1.007
B5	2000–2500	79.5–74.7	1.007–0.957
B6	2500–3000	74.7–70.1	0.957–0.909
B7	3000–3500	70.1–65.8	0.909–0.863
B8	3500–4000	65.8–61.7	0.863–0.819

). Differences between bands and clusters were assessed with Kruskal–Wallis followed by Dunn pairwise comparison with Bonferroni correction [17]. Effect sizes were quantified with Cliff’s $\delta$; altitude–emission correlations with Spearman’s $r_s$. Analysis in Python 3.11. Significance level: $\alpha =0.05$.

Because consecutive second-by-second records from the same continuous logging session are not statistically independent—engine and driving state are strongly autocorrelated over short timescales—treating all 94,263 records as independent observations in the Kruskal–Wallis framework above risks underestimating standard errors and overstating significance. Vehicle- and trip-level identifiers were not retained in the final consolidated measurement dataset (Section 4.5), which precludes a mixed-effects model with vehicle or trip as a random effect, or trip-level aggregation, as alternative corrections. We therefore validated the significance of the principal altitudinal contrasts with a non-overlapping block bootstrap: records were partitioned into contiguous 60-second blocks (1718 blocks in total, matching the timescale of short-range autocorrelation in second-by-second emissions and driving-state data), blocks were resampled with replacement (1000 replicates), and band-wise medians of $R_{\mathrm{CO}}$, $R_{\mathrm{HC}}$, $\lambda$ and ${\mathrm{EF}}_i$ were recomputed for each replicate to obtain block-bootstrap 95% confidence intervals.

The principal contrasts reported in Results retained non-overlapping block-bootstrap confidence intervals (e.g., $R_{\mathrm{CO}}$: 500–1000 m [0.0072–0.0103] vs. 2000–2500 m [0.0516–0.0806]; $\lambda$: 500–1000 m [1.235–1.349] vs. 3000–3500 m [1.070–1.147]), confirming that the reported altitudinal effect is not an artefact of pseudo-replication. Some finer between-band distinctions—particularly for ${\mathrm{EF}}_{\mathrm{NO}}$ and $R_{\mathrm{HC}}$ in bands with fewer moving-vehicle records—showed wider block-bootstrap intervals than the naive Kruskal–Wallis framework implied and should be interpreted with corresponding caution; full block-bootstrap intervals for all bands and variables are available from the corresponding author upon request.

Table 5. Records, cumulative driving duration and distance travelled per altitudinal band in the final analysis dataset (94,263 records). Vehicle- and trip-level identifiers were not retained during dataset consolidation and could not be reported by band (Section 4.5); the ten-vehicle fleet composition is reported in aggregate in Table 1.

Band	Records	Duration (h)	Distance (km)
0–500 m	21,882	6.08	185.1
500–1000 m	14,188	3.94	18.5
1000–1500 m	5729	1.59	21.6
1500–2000 m	2316	0.64	8.8
2000–2500 m	4081	1.13	14.2
2500–3000 m	22,834	6.34	26.2
3000–3500 m	16,327	4.54	77.1
3500–4000 m	6906	1.92	53.3
Total	94,263	26.18	404.9

reports the record counts, cumulative driving duration and distance travelled per altitudinal band underlying these analyses.

3. Results

3.1. Dataset Description

The final dataset comprised 94,263 second-by-second records distributed across eight altitudinal bands (

Table 6. Record distribution by altitudinal band.

Altitudinal Band	Records (n)	%
0–500 m	21,882	23.2
500–1000 m	14,188	15.1
1000–1500 m	5729	6.1
1500–2000 m	2316	2.5
2000–2500 m	4081	4.3
2500–3000 m	22,834	24.2
3000–3500 m	16,327	17.3
3500–4000 m	6906	7.3
Total	94,263	100.0

). The 2500–3000 m band concentrated the most records (22,834; 24.2%), followed by the 0–500 m band (21,882; 23.2%). Light cruise was the predominant operational mode (27,966 records; 29.7%).

3.2. Combustion Quality Indicators: $R_{\mathrm{CO}}$ and $R_{\mathrm{HC}}$

Both ratios showed statistically significant differences across bands (Kruskal–Wallis: $H=$ 13,167.90, $p<0.001$ for $R_{\mathrm{CO}}$; $H=$ 10,665.80, $p<0.001$ for $R_{\mathrm{HC}}$).

Table 7. Medians (P25–P75) of combustion quality indicators by altitudinal band.

Band	${\mathit{R}}_{\mathbf{CO}}$ = ${\mathrm{CO/CO}}_2$		${\mathit{R}}_{\mathbf{HC}}$ = ${\mathrm{HC/CO}}_2$
	Median	P25–P75	Median	P25–P75
0–500 m	0.0495	0.0098–0.0842	0.0012	0.0001–0.0021
500–1000 m	0.0086	0.0027–0.0235	0.0001	0.0001–0.0005
1000–1500 m	0.0538	0.0131–0.0919	0.0013	0.0002–0.0029
1500–2000 m	0.0491	0.0253–0.0868	0.0014	0.0002–0.0031
2000–2500 m	0.0637	0.0382–0.1123	0.0018	0.0006–0.0050
2500–3000 m	0.0221	0.0052–0.0500	0.0003	0.0002–0.0010
3000–3500 m	0.0504	0.0194–0.0835	0.0005	0.0002–0.0017
3500–4000 m	0.0643	0.0440–0.1000	0.0017	0.0003–0.0027
KW H	13,167.90 ***		10,665.80 ***

*** $p<0.001$.

presents medians and IQR by band; Figure 2 shows the full distributions.

$R_{\mathrm{CO}}$ reached its highest value in the 3500–4000 m band (median $=0.0643$), 7.5 times the value of the 500–1000 m band ($=0.0086$), with medium effect size (${\delta }_{\mathrm{Cliff}}=0.453$ between 0–500 m and 500–1000 m). $R_{\mathrm{HC}}$ in the 2000–2500 m band (0.0018) exceeded the 500–1000 m value (0.0001) by 18 times. Both indicators showed a non-monotonic pattern with relative minima in the 500–1000 m and 2500–3000 m bands.

3.3. Relationship Between Lambda and Altitude

The air–fuel ratio ($\lambda$) showed significant differences across bands ($H=1534.81$, $p<0.001$), with small effect sizes ($\left|\delta \right|\le 0.175$). Spearman correlation between altitude and $\lambda$ was negative and significant in high-demand clusters: High demand ($r_s=-0.205$, $p<0.001$) and Maximum demand ($r_s=-0.139$, $p<0.001$), indicating mixture enrichment under high load at altitude. The correlation was virtually null in Positive acceleration ($r_s=-0.001$, $p=0.870$). Figure 3 shows the $\lambda$ heatmap by band and cluster.

3.4. Volumetric Exhaust Gas Concentrations

Table 8. Median volumetric concentrations by altitudinal band. All Kruskal–Wallis contrasts significant ($p<0.001$).

Band	CO (%vol)	${\mathrm{CO}}_2$ (%vol)	HC (ppm)	NO (ppm)
0–500 m	0.52	12.00	127.0	52.7
500–1000 m	0.08	11.40	15.0	3.5
1000–1500 m	0.57	11.90	164.0	141.0
1500–2000 m	0.54	11.80	131.0	147.0
2000–2500 m	0.68	11.10	188.0	212.0
2500–3000 m	0.23	11.85	27.3	28.0
3000–3500 m	0.61	12.10	35.5	165.0
3500–4000 m	0.77	11.80	192.0	613.0
KW H	10,926 ***	1384 ***	9788 ***	17,289 ***

*** $p<0.001$.

summarises median volumetric concentrations by band. Figure 4 illustrates altitudinal trends.

Carbon monoxide (CO). Median CO increased from the absolute minimum of 0.08 %vol in the 500–1000 m band to 0.77 %vol in the 3500–4000 m band (9.6×).

Unburned hydrocarbons (HC). The 2000–2500 m band recorded the highest median HC (188 ppm), 12.5 times the minimum of the 500–1000 m band (15 ppm).

Nitric oxide (NO). NO exhibited the most pronounced non-linear behaviour ($H=\mathrm{17,289.36}$, $p<0.001$). From a minimum of 3.5 ppm in 500–1000 m, concentrations rose to 212 ppm in 2000–2500 m, then fell to 28 ppm in 2500–3000 m and rose again to the absolute maximum of 613 ppm in 3500–4000 m (Figure 5). This “N”-pattern is discussed in Section 4.

Carbon dioxide (${\mathrm{CO}}_2$). ${\mathrm{CO}}_2$ concentrations remained relatively stable (range: 11.10–12.10 %vol; 9% relative variation), confirming that the ECU maintains overall stoichiometric control.

3.5. Emission Factors by Altitudinal Band

Emission factors (EF, g/km) showed significant differences across bands for all gases (CO: $H=4866.07$; ${\mathrm{CO}}_2$: $H=5678.17$; HC: $H=6679.95$; NO: $H=$ 11,564.35; all $p<0.001$).

Table 9. Median emission factors (g/km) and fuel consumption (L/100 km) by altitudinal band.

Band	EFCO (g/km)	EFHC (g/km)	EFNO (g/km)	EF${\mathrm{CO}}_2$ (g/km)	Fuel cons. (L/100 km)
0–500 m	4.43	0.035	0.28	181.1	8.58
500–1000 m	1.32	0.008	0.01	227.9	10.01
1000–1500 m	3.89	0.039	0.16	147.6	7.97
1500–2000 m	6.12	0.066	0.23	198.2	11.56
2000–2500 m	6.42	0.068	0.25	143.1	8.83
2500–3000 m	2.51	0.012	0.05	146.4	8.45
3000–3500 m	3.60	0.015	0.14	125.4	8.16
3500–4000 m	4.41	0.034	0.36	112.7	8.43

consolidates EF medians and fuel consumption. Figure 6 disaggregates EF by K-Means cluster.

EF_CO peaked in the 2000–2500 m band (6.42 g/km), 4.9 times the minimum of the 500–1000 m band (1.32 g/km). EF_NO showed the most pronounced behaviour: from 0.01 g/km in 500–1000 m to 0.36 g/km in 3500–4000 m (36 times). EF_{${\mathrm{CO}}_2$} showed a decreasing trend with altitude, from 227.9 g/km (500–1000 m) to 112.7 g/km (3500–4000 m), consistent with reduced aerodynamic resistance at lower air density.

3.6. Effect of Road Gradient on Emission Factors

Road gradient (Section 2.7.3), tested as a continuous variable, showed only a modest, altitudinal-band-dependent monotonic association with ${\mathrm{EF}}_{{\mathrm{CO}}_2}$ (Spearman $\rho$ on $\left|\mathrm{slope}\right|$, $-0.09$ to $+0.36$ across bands). A much stronger and fully consistent pattern emerged, however, when records were instead stratified by gradient direction (

Table 10. Median ${\mathrm{EF}}_{{\mathrm{CO}}_2}$ (g/km) by altitudinal band and road-gradient category (downhill $\le -3\%$, flat $-3\%$ to $3\%$, uphill $\ge 3\%$). All within-band differences significant at $p<{10}^{-14}$ (Kruskal–Wallis).

Band	Downhill (n)	Flat (n)	Uphill (n)
0–500 m	95.96 (1794)	153.40 (6519)	159.80 (1905)
500–1000 m	53.33 (400)	81.85 (311)	219.35 (743)
1000–1500 m	74.89 (258)	121.62 (501)	215.28 (839)
1500–2000 m	52.83 (157)	182.33 (120)	219.88 (711)
2000–2500 m	76.27 (509)	164.51 (347)	196.37 (760)
2500–3000 m	51.40 (739)	104.37 (644)	150.92 (729)
3000–3500 m	42.61 (1943)	91.71 (2014)	141.90 (2153)
3500–4000 m	37.48 (1332)	82.35 (1199)	136.08 (1754)

). In every one of the eight altitudinal bands, ${\mathrm{EF}}_{{\mathrm{CO}}_2}$ was systematically lowest for downhill segments, intermediate for flat segments, and highest for uphill segments, with the difference reaching 2–4× between downhill and uphill medians in a given band. This effect was significant in every band (Kruskal–Wallis, $p<{10}^{-14}$ throughout; Table 10), consistent with the direct role of grade in tractive power demand already embedded in the ${\mathrm{PSV}}_{\mathrm{m}10}$ formulation (Equation (7)). This confirms that gradient contributes an effect on emissions that is independent of, and comparable in magnitude to, the altitudinal effect reported above, and should be controlled for—as done here—rather than omitted, when comparing emission factors across corridors with heterogeneous terrain profiles.

3.7. Fuel Consumption by Altitudinal Band

Fuel consumption showed significant differences ($H=1293.27$, $p<0.001$), with small consecutive-band effect sizes ($\left|\delta \right|\le 0.254$). Median ranged from 7.97 L/100 km (1000–1500 m) to 11.56 L/100 km (1500–2000 m). Spearman correlations revealed mode-dependent effects: High demand showed positive correlation ($r_s=+0.153$, $p<0.001$), while Light cruise ($r_s=-0.127$) and Positive acceleration ($r_s=-0.139$) showed negative correlations (both $p<0.001$).

4. Discussion

4.1. Combustion Quality Along the Altitudinal Gradient

The $R_{\mathrm{CO}}$ and $R_{\mathrm{HC}}$ ratios proved to be sensitive and standardised indicators of combustion quality deterioration with altitude. The 7.5-fold increase in $R_{\mathrm{CO}}$ between the 500–1000 m and 3500–4000 m bands exceeds the 2.56-fold factor reported by [11] for absolute CO in the 700–2400 m range in China VI vehicles, and the 2–3-fold increase reported by [7] between 0 and 3500 m in India. The superiority of $R_{\mathrm{CO}}$ as an indicator lies in that, by normalising to ${\mathrm{CO}}_2$, it eliminates confounding from displacement, exhaust mass flow and barometric dilution effects at high altitude [25]. This approach is analogously superior to the ${\mathrm{EF}}_{\mathrm{i}}^{*}$ factors proposed by [18] for diesel buses, extending it now to gasoline vehicles and continuous altitudinal gradients.

The non-monotonic pattern—with relative minima in the 500–1000 m and 2500–3000 m bands—reflects the interaction between route driving profiles and the altitude response of fuel management. The 500–1000 m band corresponds mainly to the coastal Riobamba–Guayaquil route, where the predominance of Light cruise mode at moderate speeds favours complete combustion regardless of altitude. This phenomenon is consistent with observations by [18] in diesel buses on Mexican mountain highways, where CO emission factors were significantly lower on the highway corridor than in urban areas despite higher altitude, due to a more efficient driving profile.

4.2. Non-Linear Behaviour of Nitric Oxide

The “N”-pattern of NO—minimum at 500–1000 m (3.5 ppm), local maximum at 2000–2500 m (212 ppm), descent to 28 ppm at 2500–3000 m, and absolute maximum at 3500–4000 m (613 ppm)—constitutes the most original finding of this study for three reasons. First, its magnitude: the absolute maximum of 613 ppm exceeds the minimum by 175 times and is 2.9 times the sea-level value (52.7 ppm at 0–500 m). Second, its altitudinal range: previous studies in gasoline reported the ${\mathrm{NO}}_x$ peak up to ≈1900 m in China VI vehicles [11] or up to 2000 m in diesel engines with the MEDAS simulator [8]; this study documents for the first time a second absolute maximum above 3500 m. Third, the nature of the gradient: while previous studies used static conditions or regulated cycles, the data presented here correspond to naturalistic driving, eliminating cycle bias.

The underlying mechanism can be explained by the superposition of two opposing effects. On one hand, the reduction of partial oxygen pressure decreases peak flame temperature and thereby the Zeldovich thermal NO formation rate [4]. On the other hand, progressive EGR suppression—documented by [8] from 2000 m in Euro 4 engines—increases chamber temperature by eliminating inert gas mass from the cycle, favouring NO formation. At extreme altitudes (>3000 m), the second effect outweighs the first, especially in Euro 2–3 fleet vehicles without adaptive compensation, explaining the second maximum at 3500–4000 m.

4.3. Fuel Consumption and Operational Demand

Fuel consumption did not follow a monotonic trend with altitude, in line with observations of [13] for a Euro 3 vehicle on an altimetric test bench and [14] using GT-Power up to 3030 m. The consumption minimum at 1000–1500 m (7.97 L/100 km) and maximum at 1500–2000 m (11.56 L/100 km) reflect the interaction among: aerodynamic resistance reduction with lower air density [13], progressive mixture enrichment increasing fuel flow rate, and additional mechanical demand on steep Andean gradients.

Cluster analysis revealed a key dependency: the altitude–consumption correlation changes sign between clusters. In High demand the correlation is positive ($r_s=0.153$), while in Light cruise it is negative ($r_s=-0.127$), consistent with aerodynamic resistance reduction proportional to ${\rho }_a{v}^2$ [14]. This result demonstrates that studies not controlling the operational mode produce consumption estimates that mix opposing effects and are inconsistent across corridors with different mode compositions.

4.4. Implications for High-Altitude Emission Inventories

Results have direct implications for Andean emission inventory construction. Standard EF values used in tools such as MOVES or COPERT—derived from regulated low-altitude cycles—significantly underestimate CO and HC emissions in the 1500–4000 m range. The maximum EF_CO (6.42 g/km at 2000–2500 m) is 4.9 times the coastal minimum (1.32 g/km at 500–1000 m), and [18] found similarly that CO emission factors for diesel buses in Mexico (>2000 m) were three times IPCC reference values. Using $R_{\mathrm{CO}}$ and $R_{\mathrm{HC}}$ stratified by altitudinal band and operational cluster provides a normalised, easily replicable alternative to update these inventories without exhaust mass flow measurement.

4.5. Study Limitations

The main limitations of this study are listed below.

(i)

The ten-vehicle fleet, while deliberately representing the Euro 2–5 spectrum of the Ecuadorian fleet, does not allow statistically robust stratification by Euro standards independently.

(ii)

The Brain Bee AGS-688 operates as an exhaust pipe probe rather than a dilution PEMS, excluding absolute mass emission calculations with the precision of equipment such as the Semtech ECOSTAR [18].

(iii)

Driving routes are limited to five Ecuadorian corridors.

(iv)

Ambient temperature effects—which co-vary with altitude in Andean corridors—were not explicitly decoupled in the statistical analysis.

(v)

Above ≈1500 m a.s.l., ambient pressure falls below the analyser’s certified automatic compensation range (85.0–106.0 kPa). This introduces a potential systematic bias in raw concentration readings that was not independently verified with certified span gases at altitude.

(vi)

Emission factors reported in Section 3 use a nominal stoichiometric air–fuel ratio and do not include the explicit wet-to-dry correction derived in Section 2.4. A validation check against the fully specified calculation (Section 2.4) shows a modest, altitude-independent margin (median $-10\%$ to $-6\%$), indicating that absolute EF magnitudes carry an estimated ±10% uncertainty from this source, while the altitudinal trends themselves are not affected.

(vii)

Vehicle- and trip-level identifiers were not retained in the final consolidated measurement dataset, so records cannot be attributed to individual vehicles or driving sessions post hoc. This precluded a mixed-effects model with vehicle or trip as a random effect, and prevented reporting the number of vehicles, trips or distance travelled by individual vehicles within each altitudinal band. Cumulative duration and distance by band, which do not require this attribution, are reported instead in Table 5. The significance of the principal altitudinal contrasts was independently confirmed with a block-bootstrap procedure that does not require vehicle or trip identifiers (Section 2.9). As further, complementary evidence that the reported altitude effects are not primarily an artefact of fleet composition, a related analysis of the same instrumented fleet (Grupo Tipológico Vehicular, GTV [31]) grouped vehicles into five typological clusters by specific fuel-consumption efficiency. That analysis found the altitude-related increase in EF_CO to be consistent in direction across all five clusters, rather than confined to a subset of vehicle types. The GTV comparison uses a driving-window-matched subsample rather than the full dataset analysed here, so it is reported here as corroborating, not confirmatory, evidence.

(viii)

No independent manufacturer accuracy specification for the electrochemical NO cell was available, since OIML R99 does not define a certified accuracy class for this channel (Table 2). Reported NO values should therefore be interpreted with the instrument’s resolution (1 ppmvol) and response time as the primary independently verified performance figures. Field cross-validation from a companion campaign using the same instrument class [31] found the electrochemical NO cell to agree with a reference PEMS within ±30–50 ppm above 2500 m a.s.l., within the range admissible for portable electrochemical sensors under ISO 16183 [32]. We report this as supporting context rather than as a calibration record specific to the instrument units used in this study (Table 2).

5. Implications for Carbon and Pollutant Reduction in Combustion Applications

The findings of this study carry direct implications for ongoing efforts to reduce carbon, ${\mathrm{NO}}_x$ and ${\mathrm{SO}}_x$-related emissions from existing and emerging combustion applications, particularly in regions where engines and combustion systems operate under variable air density. For the spark-ignition gasoline platform studied here, ${\mathrm{SO}}_x$ formation is governed primarily by fuel sulfur content, rather than by the altitude-driven combustion changes documented in this work. The ${\mathrm{CO/CO}}_2$- and ${\mathrm{HC/CO}}_2$-based diagnostic framework therefore targets carbon-efficiency and ${\mathrm{NO}}_x$-relevant combustion quality directly. Its relevance to ${\mathrm{SO}}_x$ mitigation is indirect: it acts through the same fuel-quality and combustion-control interventions, discussed below, that also constrain ${\mathrm{NO}}_x$ and incomplete-combustion products. These interventions apply most directly to combustion systems burning higher-sulfur fuels, such as the marine and stationary diesel/heavy-fuel applications discussed below.

5.1. Altitude as an Overlooked Variable in Carbon Reduction Strategies

Most decarbonisation roadmaps for the light-duty fleet—fuel reformulation, hybridisation, or fleet renewal schedules—are calibrated using emission factors derived from sea-level or near-sea-level test cycles. Our results show that EF_{${\mathrm{CO}}_2$} is not the limiting concern at altitude; rather, it decreases moderately with elevation due to reduced aerodynamic load (Table 9). The critical risk lies instead in the substantial underestimation of CO and HC emission factors above ≈2000 m a.s.l., where EF_CO reaches 6.42 g/km—4.9 times the coastal reference value. Carbon-reduction policies that rely on low-altitude EF values to project co-benefits in criteria pollutant reduction will therefore systematically underestimate the air-quality burden of the existing gasoline fleet in Andean and other high-altitude regions, weakening the evidence base for prioritising fleet renewal or retrofit programmes in these areas.

5.2. A Critical Altitude Threshold for Combustion Control Intervention

The pronounced increase in $R_{\mathrm{CO}}$ and $R_{\mathrm{HC}}$ beyond 2000 m, and the divergence of the NO “N”-pattern above 3000 m, identify ≈2000 m a.s.l. (≈79 kPa) as the altitude at which conventional spark-ignition combustion control—particularly in Euro 2–3 vehicles without adaptive fuelling or EGR modulation—loses effectiveness in maintaining low-emission combustion. This threshold provides a quantitative target for prioritising interventions such as: (i) adaptive air–fuel ratio control recalibrated for barometric pressure rather than fixed look-up tables; (ii) EGR systems designed to remain partially active at high altitude rather than fully closing near 2000 m, to limit the NO rebound documented at 3500–4000 m; and (iii) accelerated electrification or hybridisation incentives specifically targeted at fleets operating above this threshold, where the combustion-based emission penalty is largest and the relative benefit of electrified powertrains is correspondingly highest.

5.3. Transferability to Broader Combustion Systems

While this study focuses on light-duty gasoline engines, the underlying physical driver—reduced oxidant availability and altered thermal conditions at lower ambient pressure—is not unique to road vehicles. The $R_{\mathrm{CO}}$/$R_{\mathrm{HC}}$ framework, normalised to ${\mathrm{CO}}_2$ and therefore independent of fuel flow or dilution, is in principle transferable to any combustion system operating under variable air density or oxidant supply, including stationary generators at high elevation, dual-fuel and marine engines subject to variable intake conditions, and combustion-based heating systems in high-altitude communities. In all these applications, the same diagnostic logic applies: monitoring ${\mathrm{CO/CO}}_2$ and ${\mathrm{HC/CO}}_2$ ratios offers a low-cost, mass-flow-independent indicator of combustion quality degradation that can flag when a system is operating outside its designed efficiency envelope—supporting earlier and more targeted interventions to reduce carbon and criteria pollutant emissions across the broader landscape of combustion applications addressed by this Special Issue.

6. Conclusions

This study presents the first naturalistic evidence at the Andean corridor scale of the effect of altitude on combustion efficiency and emissions of light-duty gasoline vehicles in the 0–4000 m a.s.l. range, based on 94,263 second-by-second records from a mixed Euro 2–5 fleet in Ecuador. The main conclusions are:

• The ratios $R_{\mathrm{CO}}$ and $R_{\mathrm{HC}}$ are effective standardised indicators of combustion quality at altitude. $R_{\mathrm{CO}}$ increased 7.5 times between the coastal 500–1000 m band and the 3500–4000 m band, and $R_{\mathrm{HC}}$ by 18 times. Their implementation does not require exhaust mass flow measurement, making them accessible for fleet monitoring in middle-income countries.
• Nitric oxide exhibits a non-linear “N”-pattern with an absolute maximum at 3500–4000 m. Median NO concentration reached 613 ppm—175 times the minimum of 3.5 ppm at 500–1000 m—as a result of the interplay between EGR suppression and reduction of partial oxygen pressure. This pattern, documented here for the first time in gasoline over a continuous 4000 m gradient, implies that low-altitude ${\mathrm{NO}}_x$ limits are insufficient to characterise real impact in high mountain areas.
• The operational mode regulates the intensity of the altitudinal effect on consumption and emissions. The altitude–consumption correlation changed sign between High demand ($r_s=+0.153$) and Light cruise ($r_s=-0.127$), and the altitude–$\lambda$ correlation was virtually null in Positive acceleration ($r_s=-0.001$, $p=0.870$) but significant in High demand ($r_s=-0.205$, $p<0.001$). This demonstrates that studies not controlling operational demand produce ambiguous estimates of the altitudinal effect.
• CO emission factors in the 2000–2500 m band are 4.9 times the coastal values. EF_CO reached 6.42 g/km at 2000–2500 m versus 1.32 g/km at 500–1000 m. At this threshold of ≈2000 m a.s.l., pressure drops to ≈79 kPa. Beyond this point, Euro 2–3 engines without adaptive compensation lose stoichiometric control, so the threshold should be considered as an operational limit in national regulations of Andean countries.
• Fuel consumption does not follow a monotonic trend with altitude. Variation between 7.97 L/100 km (1000–1500 m) and 11.56 L/100 km (1500–2000 m) reflects competition between aerodynamic resistance reduction at lower air density and increased fuel expenditure to compensate engine performance loss. This confirms the findings of [13,14] in bench and model conditions, and extends them to the Andean naturalistic context.

These results provide the first naturalistic evidence base for the calibration of emission inventory models in Andean corridors and for informing vehicle fleet renewal policies in countries with high altitudinal exposure.

Recommendations for policy and practice.

(i)

The ≈2000 m a.s.l. threshold identified for CO (Conclusion 4) should be considered explicitly in periodic vehicle inspection limits and fleet-renewal incentives in Andean countries, where uniform sea-level-calibrated thresholds systematically under-detect high-altitude non-compliance in Euro 2–3 vehicles.

(ii)

Emission inventory models currently calibrated with sea-level or near-sea-level factors should incorporate altitude- and gradient-specific correction factors of the type reported in Table 9 and Table 10, rather than assuming spatially uniform emission factors, when applied to mountainous urban regions.

(iii)

Road gradient was found to modulate ${\mathrm{EF}}_{{\mathrm{CO}}_2}$ by a margin comparable to the altitude effect itself (Section 3.6). For this reason, route-level emissions inventories in mountainous terrain should stratify by gradient in addition to altitude.

Recommendations for future research.

(i)

Field verification of gas-analyser performance with certified span gases above ≈1500 m a.s.l. is needed to directly quantify the compensation-range uncertainty identified in Section 2.3.1, rather than bounding it indirectly.

(ii)

Future data-collection protocols should retain vehicle- and trip-level identifiers throughout consolidation. This would enable mixed-effects modelling of altitude effects, rather than the block-bootstrap validation used here as a second-best alternative (Section 2.9).

(iii)

Ambient temperature and altitude co-vary in Andean corridors. A factorial or climate-controlled design is needed to decouple their independent contributions to the reported effects.

(iv)

The $R_{\mathrm{CO}}$/$R_{\mathrm{HC}}$ indicators and the $\lambda$-corrected, wet-to-dry mass-balance approach developed in Section 2.4 should be extended to diesel, hybrid and natural-gas vehicles, and to larger fleets that allow stratification by Euro standard.

(v)

Instrument-specific cross-validation of electrochemical NO-sensor accuracy against a reference PEMS is needed for the exact units deployed in a given campaign. A companion campaign using the same instrument class reported ±30–50 ppm agreement above 2500 m a.s.l. against a reference PEMS [31], but no certified accuracy class currently exists for this channel under OIML R99 (Table 2), and this figure was not obtained for the specific instrument units used in the present study.