Effect size of COVID-19 mobility restrictions intensity on air pollution: a natural experiment Authors: Aníbal A Teherán, Luis M Pombo, Gabriel Camero-Ramos, Jaime A Robayo, William J Rodríguez, Juan C Correa, Karen P Ayala Published: 2026 ABSTRACT AbstractTo improve air quality, regulatory actions targeting road traffic and industrial emissions must be implemented. This study determined the effect size (ES, Cohen’s δ) and the significance of progressive quarantine restrictions levels national quarantine [NQ+], smart quarantine [SQ++], and focalized quarantine [FQ+++] on air pollution in Bogotá, Colombia, during the COVID-19 pandemic. A natural experiment showed that, during SQ and FQ phases, concentrations of [PM₁₀, 20.0 µg/m3; PM₂.₅, 8.03 µg/m3; SO2, 1.61 µg/m3] and [PM₁₀, 23.2 µg/m3, PM₂.₅, 8.84 µg/m3], respectively, were significantly lower than during equivalent pre-pandemic periods (year 2019; p < 0.001). Effect sizes were larger during FQ (PM₁₀: -1.220, PM₂.₅: -0.605) than during SQ (PM₁₀: -1.180, PM₂.₅: -0.252, SO2: -0.388). Although pollutant concentrations remained below regulatory thresholds (environmentally significance), the ES for [PM₁₀ µg/m3; PM₂.₅ µg/m3] were larger during SQ and FQ. These findings support public health strategies aimed at strengthening the control of emission sources that degrade air quality, particularly those increasing [PM₁₀, PM₂.₅], Moreover, the introduction of the concept of “environmentally significant” effects may further improve decision-making and address environmental problems that affect human well-being.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1038/s41598-026-54722-7.Keywords: Social interaction, Quarantine, COVID-19, Air pollution, Linear models, Non-randomized controlled trials as topicSubject terms: Climate sciences, Environmental sciences, Environmental social sciences Introduction Restrictions on social interaction during the COVID-19 pandemic, including restrictions on vehicle mobility, had a favorable effect on air pollution, resulting in a substantial reduction in nitrogen dioxide [NO2], sulfur dioxide [SO2] and carbon monoxide [CO]1,2. However, globally, ambiguous effects were observed in major cities for particulate matter [PM] and ozone [O3], possibly due to the varying intensity of restrictions or habitual social dynamics that resurfaced with subsequent epidemic waves2–4.. The COVID-19 pandemic served as an ideal natural scenario for determining the effect on air pollutants resulting from anthropogenic causes (x1, x2, x3… xk)4–6. Previous research prior to COVID-19 had limitations in establishing a causal relationship between these variables, mainly because restricting social interaction and vehicular traffic has a significant negative impact on the financial dimension and various economic sectors4,7,8. In Bogotá, at least two orange alerts are issued every year due to excessive concentrations of air pollutants, which have also been associated with increases in healthcare visits due to exacerbations of respiratory diseases9,10. The problem is so significant that, compared to other large cities, the most favorable results in terms of air quality were observed in Bogotá during the COVID-19 quarantine (COVID-Q)11,12. Due to the increase in transmissibility, morbidity, and mortality during the course of the COVID-19 pandemic, non-pharmacological preventive measures were intensified in Bogotá, especially the COVID-Q.1313. Three types of quarantine were implemented, national quarantine (NQ), smart quarantine (SQ), and focalized quarantine (FQ), with gradually increasing restrictions on citizen mobility. These measures had a favorable impact and may have had significant effects on air pollution. It is worth noting that the analysis of air quality during the COVID-19 pandemic period has provided a window of opportunity to implement measures that, incidentally, improved air quality, including mobility restrictions14. The effects and impact of COVID-Q on air pollution have been investigated using descriptive, ecological, and cross-sectional designs, as well as various review types, applying descriptive statistics, geospatial analysis, time series models, and machine learning techniques1,2,12,13. However, given the abrupt disruption to citizen mobility during the pandemic and the staggered intensity of COVID-Q, it was the ideal for estimating effects with rapid manifestations and for determining a gradual magnitude of effect (moderate to large), which are key components for resolving a hypothesis through a natural experiment15. To mitigate the negative anthropogenic impacts on air quality, a range of measures have been implemented, including temporary vehicle circulation restrictions, congestion management strategies, car-free days, traffic demand management during peak hours, the establishment of low-emission zones in high-traffic corridors, and freight scheduling restrictions in urban centers. However, uncertainty remains regarding the appropriate intensity and calibration of these interventions. Although this study relies on data from an exceptional period (the pandemic), the findings may provide valuable insights for refining and optimizing the design and implementation of such strategies. To evaluate the favorable effects of mobility restriction intensity on air pollution, the COVID-19 pandemic was used as a practical case study. This research was designed to address three objectives: (1) to calculate the effect on air pollutants for the three levels of COVID-Q, (2) to determine the effect of the time of measurement during the day on air pollution, and (3) to establish whether the effects of COVID-Q on air pollutants were environmentally significant. Methods Setting, place, and registers A natural quasi-experiment was conducted using data from the Bogotá Air Quality Monitoring Network (RMCAB; open-access data). Hourly measurements of air pollutants and meteorological variables from 2019 to 2020 were included, obtained from the monitoring station located in the Suba district. The study was reviewed and approved by the Research Center of the Fundación Universitaria Juan N. Corpas. As this study exclusively utilized publicly available, anonymized, and aggregated environmental data (open data), it was classified as “No Risk” research, in compliance with standard ethical guidelines for epidemiological and environmental studies. Database and variables Information was extracted from three quarantine periods (NQ, SQ, FQ) from the 2020 database and from symmetrical periods in the 2019 database. Each period contained measurements taken over 24 h, one week before the start of the quarantines (t0) and four days after (t1[24 horas], t2[48 horas], t3[72 horas], t4[96 horas]). The variables analyzed were the concentration of [PM₂.₅ µg/m3], [PM₁₀ µg/m3], [SO2 µg/m3], [O3 µg/m3], ambient temperature (°C), wind speed (Km/h), wind direction ° (clockwise) and time of the day (0700–1859 h[day], 1900–0659 h[night]). Statistical analysis Data analysis was performed using the R programming language and the JASP statistical software (Version 0.17.1.0). A sample size of 595 record was calculated to estimate the air pollutants explained by 12 potential predictors (COVID-Q, follow-up times, time of the day, comparison years), 1 − β: 0.9, α: 0.0041 (Bonferroni adjustment, 0.05/12 predictors) y and effect size 0.25 (G*Power 3.1.9.7. software). In the quasi-experimental design, the COVID-Q periods in 2020 were established as the treatment groups for air pollutants, and the symmetrical periods in 2019 served as control groups. Data were reported as means and standard deviations (SD), and the correlation between the air pollutants metric and environmental variables (rho, ρ) was assessed to determine their inclusion in the multivariate model. To estimate differences in air pollutants (IC95%) the one-tailed t-test was applied (H1 = [air pollutionCOVID−Q] – [air pollutionControl] < 0; x̄ differences effect size Cohen´s δ). This test was also used to compare air pollution between COVID-Q and control periods, during the day or night, and to compare, individually, the air pollution of each post-quarantine time with the pre-quarantine. Additionally, a one-way ANOVA (Bonferroni post-hoc) was carried out to determine the exhaustion of the effect of quarantines between times t1[24 h] y t4[96 h]. To evaluate the effect on air pollution associated with the potential interaction between each COVID-Q period [x1, dummy: 1/0referent] and the eight daytime measurements [x2, dummy: 1/0referent], (x3, COVID-Q*día), 12 generalized liner models (GLMs), were applied, one for each pairing between air pollutant ~ COVID-Q, for example: ŷ1: x̄[PM₁₀] ~ β0 + β1(COVID NQ [1]/Control NQ [0] * β2(Day [1]/Night [0] + ε) ŷ1: x̄[PM₁₀] ~ β0 + β1(COVID NQ [1]/Control NQ [0] * β2(Day [1]/Night [0] + ε) ŷ1: x̄[PM₁₀] ~ β0 + β1(COVID SQ [1]/Control SQ [0] * β2(Day [1]/Night [0] + ε) ŷ1: x̄[PM₁₀] ~ β0 + β1(COVID SQ [1]/Control SQ [0] * β2(Day [1]/Night [0] + ε) ŷ1: x̄[PM₁₀] ~ β0 + β1(COVID FQ [1]/Control FQ [0] * β2(Day [1]/Night [0] + ε) ŷ1: x̄[PM₁₀] ~ β0 + β1(COVID FQ [1]/Control FQ [0] * β2(Day [1]/Night [0] + ε) Finally, to determine whether the decrease in air pollutants was environmentally significant, the difference between air pollution was compared with the maximum environmental threshold established for each air pollutant (H1, air pollution - environmental threshold < 0) (One tail, t-Test). Results General results For each air pollutant, 720 records were selected and evenly distributed across the quarantine periods (2020) and the corresponding control periods (2019), as well as times of day, and the pre- and post-quarantine times (t0-t4). In total, 2,880 records were analyzed, and data completeness was ≥ 95% for all variables included. The concentrations of the air pollutants PM₁₀, PM₂.₅, O₃, and SO₂ ranged from 5 to 173 µg/m³, 1–95 µg/m³, 0.40–65.7 µg/m³, and 0.20–30.6 µg/m³, respectively. Figure 1 shows the behavior of each pollutant over the 24-hour cycle during the COVID-Q periods and their corresponding control periods in 2019. During the least restrictive quarantine (NQ), PM₂.₅ and O₃ concentrations were higher than in the control periods, whereas PM₁₀ and SO₂ levels were similar in both periods. In contrast, during the more restrictive quarantines (SQ and FQ), PM₂.₅ and PM₁₀ concentrations were lower than those observed in the controls. Finally, O₃ and SO₂ concentrations showed slight reductions during daytime hours in the FQ and SQ periods, respectively. Concentration of air pollutants during the day. NQ: national quarantine, SQ: Smart quarantine, FQ: focused quarantine. Control: measurements collected during 2019. Among the environmental variables analyzed as potential modifiers of air pollution levels, temperature (°C) exhibited the strongest direct correlation with at least one of the air pollutants (O3 ≈ NQ, SQ; rho, ρ: >0.7). On the other hand, wind speed (km/h) showed the strongest indirect correlation with at least one of the air pollutants (PM₂.₅ ≈ SQ; rho, ρ: −0.4, −0.5) (Fig. 2). Correlation between air pollutants and environmental variables. NQ: national quarantine, SQ: Smart quarantine, FQ: focused quarantine. Effect on air pollutants according to the intensity of COVID-Q Table 1 presents the annual means (2020 and 2019) for each air pollutant indicator, where statistically significant differences were observed only for [PM₁₀]. The table also shows the average pollutant levels during the COVID-Q periods and their corresponding controls. Among the periods with statistical differences, the x̄ of air pollutants were lower in the more restrictive COVID-Q periods. (SQ: [PM₁₀], [PM₂.₅], [SO2]; FQ: [PM₁₀], [PM₂.₅]), however, the largest effect sizes (most favorable effects) were calculated in the most restrictive COVID-Q (FQ). Table 1Average concentration of air pollutants in 2020 Vs. 2019.Air pollutants20202019†P-valueEffect sizex̄ (SD)x̄ (SD)Left tail(Cohen´s δ)PM₁₀ 34.5 (23.7) 45.2 (25.1) < 0.001 NQ62.6 (20.8)57.0 (30.5)0.947Not apply SQ 20.0 (8.49) 34.2 (14.7) < 0.001 −1.180 (+++) FQ 23.2 (10.6) 44.3 (22.1) < 0.001 −1.220 (+++)PM₂.₅19.0 (17.6)17.9 (15.4)0.811 NQ40.0 (14.1)30.4 (18.1)1.000Not apply SQ 8.03 (5.55) 9.60 (6.85) 0.029 −0.252 (+) FQ 8.84 (5.27) 13.6 (9.86) < 0.001 −0.605 (++)O318.8 (13.8)15.2 (11.6)0.999 NQ30.3 (15.7)22.2 (13.6)1.000Not apply SQ10.7 (6.90)7.46 (6.26)0.999Not apply FQ15.3 (8.47)16.0 (8.75)0.250Not applySO21.82 (2.24)2.08 (3.01)0.095 NQ2.89 (3.24)3.20 (4.89)0.279Not apply SQ 1.61 (1.48) 2.08 (0.81) 0.001 −0.388 (++) FQ0.99 (0.83)0.99 (0.76)0.493Not applyNQ, National Quarantine; SQ, Smart Quarantine, FQ, Focused Quarantine. † COVID-Q control period. The bold p-values highlight comparisons with significant differences. Effect: negative, indicating a decrease in air pollut in 2020 compared to 2019. Effect size: large (+++), moderate (++), small or insignificant (+). Average concentration of air pollutants in 2020 Vs. 2019. NQ, National Quarantine; SQ, Smart Quarantine, FQ, Focused Quarantine. † COVID-Q control period. The bold p-values highlight comparisons with significant differences. Effect: negative, indicating a decrease in air pollut in 2020 compared to 2019. Effect size: large (+++), moderate (++), small or insignificant (+). Regarding the attenuation of the COVID-Q effect, statistical significant differences were identified during the NQ for the [PM₁₀] y [PM₂.₅], between the pre-quarantine period (t0) and at least one of the post-quarantine times t1, t2, t3, t4 (one-way ANOVA, P-Value < 0.001); Likewise, statistical differences were identified in SQ and FQ, respectively, for the [PM₁₀] y [PM₂.₅]. However, only in the NQ for the [PM₁₀] and [PM₂.₅] a pattern of effect attrition was observed (data not shown). Interaction between COVID-Q and the time of day of registration In general, PM₁₀ and O₃ concentrations were higher during daytime hours (p < 0.001). It was also observed that PM₁₀ levels were lower during both daytime and nighttime in the SQ and FQ periods. Similarly, during the FQ period, PM₂.₅ levels were reduced. In contrast, SO₂ concentrations were lower only during daytime in the SQ period (Table S1). Three types of Generalized Linear Models (GLM) were compared: Gaussian::Identity, Gaussian:: Log, and Gamma::Inverse, and based on parsimony (AIC), goodness of fit (degrees of freedom), fit (“envelope”), and least overdispersion (deviance) to estimate the interaction between COVID-Q and the time of day of recording, the following was selected Gamma::Inverso GLM (data not shown). The GLMs showed that the x̄ of the [PM₁₀], [PM₂.₅] and [SO2] were lower during the day in the SQ and FQ periods (2020/Day) compared to the controls/references (2019/Nights). Conversely, during the NQ, the x̄ values for [PM₁₀] and [PM₂.₅] were higher than the controls/references, as were the x̄ values for [O3] in all COVID-Q periods. Environmentally favorable interactions (negative estimates) were estimated during the SQ and FQ for [PM₁₀], and unfavorable interactions (positive estimates) for [PM₁₀] and [O3], respectively, in the NQ and SQ (less restrictive COVID-Q) (Table 2). Table 2Interaction model between COVID-Q and time of measurement.Air pollutants2020-Day2019 ref. – Night ref.Estimate interactionx̄ [µg/m3]x̄ [µg/m3]2020 * Day [µg/m3] (IC95%)PM₁₀ NQ 50.4 60.5 189.0 (109.9; 671.2) SQ 32.3 22.9 −89.1 (−430.6; −49.6) FQ 36.4 29.1 −66.1 (−134.0; −43.9) PM₂.₅ NQ29.737.0Non significant SQ10.56.51Non significant FQ14.27.55Non significantO3 NQ16.339.1Non significant SQ 3.00 15.4 6.41 (4.41; 11.7) FQ12.419.7Non significantSO2 NQ3.402.70Non significant SQ2.061.41Non significant FQ1.060.92Non significantNQ, National Quarantine; SQ, Smart Quarantine; FQ, Focused Quarantine. The estimated linearized concentrations (Gamma::Inverse GLM, 1/estimated parameter) that demonstrated statistical significance in the interaction model between COVID-Q and time of day are presented in bold. ref.: reference. Interaction model between COVID-Q and time of measurement. NQ, National Quarantine; SQ, Smart Quarantine; FQ, Focused Quarantine. The estimated linearized concentrations (Gamma::Inverse GLM, 1/estimated parameter) that demonstrated statistical significance in the interaction model between COVID-Q and time of day are presented in bold. ref.: reference. Except for the x̄ of [PM₂.₅] in the NQ, in all COVID-Q and control periods, the calculated x̄ of air pollution were environmentally significant (≤ maximum permitted threshold). The calculated effect size of the difference between the x̄ of [PM₁₀] and [PM₂.₅] and the maximum permitted thresholds were greater during the SQ and FQ; likewise, for the x̄ of [O3] and [SO2], respectively, in the FQ and NQ (Table 3). It should be noted that, in 2020, the largest effect sizes were calculated in the most restrictive COVID-Qs (SQ, FQ). Table 3Environmentally significant effects during the COVID-Q and control periods.Air pollutants (upper threshold)20202019H1 = x̄ - threshold < 0H1 = x̄ - threshold < 0P-value; effectP-value; effectPM₁₀ (75 µg/m3) NQ< 0.001; −0.592 (++)< 0.001; −0.589 (++) SQ< 0.001; −6.47 (+++)< 0.001; −2.78 (+++) FQ< 0.001; −4.88 (+++)< 0.001; −1.38 (+++)PM₂.₅ (37 µg/m3) NQ0.988; 0.215< 0.001; −0.363 (++) SQ< 0.001; −5.22 (+++)< 0.001; −4.00 (+++) FQ< 0.001; −5.34 (+++)< 0.001; −2.37 (+++)O3 (100 µg/m3) NQ< 0.001; −4.43 (+++)< 0.001; −5.69 (+++) SQ< 0.001; -12.8 (+++)< 0.001; −14.8 (+++) FQ< 0.001; −9.99 (+++)< 0.001; −9.59 (+++)SO2 (100 µg/m3) NQ< 0.001; −29.9 (+++)< 0.001; −19.8 (+++) SQ< 0.001; −66.3 (+++)< 0.001; −121 (+++) FQ< 0.001; -119 (+++)< 0.001; −129 (+++)Cohen’s d estimator was used to calculate the effect size of the difference between air pollut and the maximum environmental threshold established for each air pollutant. The largest calculated effect sizes are highlighted in bold. Effect size: large (+++), moderate (++), small or insignificant (+). Environmentally significant effects during the COVID-Q and control periods. Cohen’s d estimator was used to calculate the effect size of the difference between air pollut and the maximum environmental threshold established for each air pollutant. The largest calculated effect sizes are highlighted in bold. Effect size: large (+++), moderate (++), small or insignificant (+). Discussion The natural experimental design confirmed the favorable effects of the COVID-Q phases on air pollution, thereby strengthening the consistency criterion for causal inference15. Moreover, contrary to much of the available scientific evidence, this study was able to demonstrate the impact of progressively increasing COVID-Q intensity on air pollution. Bogotá is divided into 20 administrative districts, and this study focused on Suba, which has the greatest number of arterial roadway kilometers per lane and the largest population (Figure S1). Notably, the city’s vehicle fleet increased by more than 200% between 2007 and 2015, and Suba is characterized by a predominance of motorcycles, private vehicles, and mass-transport units. These characteristics, together with the district’s commercial activity profile, may have made Suba more sensitive to the gradual mobility restrictions implemented during the pandemic (COVID-Q)16,17. To improve a city’s air quality, an ideal measure would be to restrict or limit its operational activities (vehicles, industries, and others). The progression of the pandemic and the predictive modeling efforts guided the optimal intensity of non-pharmaceutical interventions, indicating that aggressive and moderately prolonged lockdowns were the most effective strategy for slowing the basic reproduction number (R₀), and for favorably influencing other epidemiological and socioeconomic outcomes18,19. Although the evidence consistently demonstrated the “impact or effect” of COVID-Q phases on air pollution, this scenario offered a unique opportunity to elucidate the effects of COVID-Q measures implemented with varying intensities15. The methodology applied in this study confirmed that the more intensive mobility restrictions (COVID-Q) amplified the reductions observed in PM₁₀, PM₂.₅, and SO₂ concentrations (FQ, SQ versus NQ). Furthermore, although the primary objective of similar publications on this topic has been to quantify the “effect” or “impact” on air pollution, none of these studies calculated or estimated effect size using a valid statistical parameter. In contrast, the present investigation employed Cohen’s δ, which, beyond a simple comparison between two statistical parameters, allowed the determination of the magnitude of the differences between them1–3,7,12,13,19,20. The generalized linear models (GLMs) revealed a favorable environmental synergy for PM₁₀ during daytime measurements in the higher-intensity quarantine phases (FQ and SQ). Given that the diesel vehicle fleet represents the primary source of this pollutant in Suba, as documented by studies conducted both before and during the pandemic, these findings reinforce the direct relationship between decreased vehicular traffic and improvements in air quality1,10,21–23. In urban environments, reductions in traffic-related emissions are often accompanied by decreases in nitrogen oxides (NOx), particularly nitric oxide (NO), which plays a central role in ozone removal through titration. Under typical conditions, NO reacts with O₃ and limits its accumulation; however, when NO emissions are reduced, this scavenging effect is weakened, potentially allowing O₃ concentrations to increase, especially under favorable photochemical conditions such as high solar radiation, as has been described for Bogotá. Therefore, the increases in [O₃] observed during some restriction phases may reflect a combination of reduced NO titration and enhanced photochemical production. It is important to note that this site-specific analysis did not include NOₓ or volatile organic compounds (VOCs), which are key to fully characterizing ozone formation regimes. Consequently, this explanation should be interpreted as a plausible mechanism supported by previous evidence, rather than a definitive causal attribution24,25. A particularly relevant finding was the environmentally significant reduction in PM₁₀, PM₂.₅, and O₃ during the more restrictive quarantine phases. The decrease in air pollution below the maximum permissible thresholds has been described as a primary outcome in previous studies; however, this is the first time that the concept of “environmentally significant differences” has been applied. The intent is to promote regulatory decisions that generate real environmental impact, rather than relying solely on statistical differences (P-value < 0.05), given that air quality improvement targets are defined based on reducing air pollutants to maximum thresholds, rather than on statistically significant differences. Furthermore, favorable effects should not be supported solely by reductions in air pollutants that do not meet international targets (maximum thresholds), and in some cases, even locally established goals. The introduction of the “environmentally significant” concept is relevant because it enables the monitoring and evaluation of the impact of public policies aimed at improving air quality10,24,26,27. Additionally, similar to what has been established in the clinical context, the application of the concept of “environmentally significant differences” is relevant for sample size calculation and statistical power, as well as for the design of future natural experiments or research studies in this field1,2,4,5,19. It is worth noting that this represents a consistent limitation in previously published studies, which was addressed in clinical research through the incorporation of the “minimally significant difference” parameter into sample size calculation formulas. This investigation shows several strengths that supported the resolution of the three proposed objectives while avoiding collateral economic and financial drawbacks. The gradual implementation of COVID-Q phases in Bogotá enabled the exploration of the dose–response criterion within a natural experiment design, with high statistical power and an analytical strategy specifically selected to minimize confounding bias (GLMs)15. Among the limitations, the findings of this study are confined to the locality of Suba, with the previously described urban and social characteristics, which partially limits the extrapolation of the results to other geographic settings. Also, the quasi-experimental nature of the design is notable, as it prevented the random allocation of measurements. Nevertheless, all consecutive records from the selected periods were included, thereby reducing selection bias. Additionally, interactions with complex meteorological variables such as the photochemical reactions between NOx and O₃ were not modeled due to instrumentation constraints at the monitoring station. (5,27)It should be noted that interannual meteorological variability in the reference year is also an additional source of uncertainty in the estimated effect sizes. Finally, the socioeconomic challenges faced by Bogotá during the COVID-Q periods may have influenced the premature return to social activities, even in violation of mobility restrictions, a condition that could have affected the expected results regarding the dissipation of COVID-Q effects (data not shown)28. Conclusions Higher levels of mobility restrictions (COVID-Q), particularly in an area with substantial vehicular flow, produced more favorable effects on air quality indicators, mainly [PM₁₀, PM₂.₅], with evidence of daytime synergy for [PM₁₀]. Moreover, during the most restrictive COVID-Q phases, “environmentally significant differences” were observed for [PM₁₀, PM₂.₅, O₃]. The hypotheses addressed in this investigation demonstrated the potential positive impact of public health strategies aimed at intensifying the control of sources that deteriorate air quality, particularly those contributing to increases in [PM₁₀, PM₂.₅]. Likewise, the concepts of “Environmentally Significant” and “Environmentally Significant Differences” should be further evaluated in future studies to determine their potential role in methodological development and in informing benefit assessment and decision-making processes in environmental health29. Research carried out during the pandemic contributed to updating and modifying existing regulations, including the shift from traditional isolated measures (e.g., “no-drive days”) to combined strategies; the transition from policies focused exclusively on transportation toward stricter regulation of area sources or integrated approaches; and providing a technical basis for the European Commission to propose the revision of the Ambient Air Quality Directive14,30. Future research could apply this methodology to evaluate the impact of environmental interventions such as car- and motorcycle-free days, license plate-based mobility restrictions, and similar measures. Moreover, public health measures that have been questioned, such as car- and motorcycle-free days, could be optimized through synergistic interventions that on the same day may further improve air quality. Examples include: (1) controlling emissions from construction activities, implementing alternate-day circulation plans, and restricting heavy-duty trucks; (2) traffic restriction policies, massive public transportation subsidies, high-penalty policies targeting illegal motor vehicle travel, and the promotion and subsidization of low-emission vehicles; and (3) same-day combinations of the measures described above14,30,31. Additional studies using similar data sources and information in other localities of Bogotá or in different cities may help establish the consistency of the observed results. Likewise, it would be important to quantify the effect size of the environmental benefits observed during the COVID-Q periods in areas with industrial or commercial activity, as well as in regions with less urbanized characteristics. This would help overcome the limited extrapolation associated with the specific geographic area in which this study was conducted (Suba, Bogotá). Moreover, future work may determine whether the estimated effects remain consistent for other air pollutants (NOx, CO) or when additional environmental variables are incorporated into the statistical models. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements We thank the Fundación Universitaria Juan N. Corpas Research Center, and the Manuela Beltrán University Master’s Program in Biostatistics, for their institutional support, academic contributions, and methodological assistance. Author contributions Conceptualization: TAA, GCR, WJR, AKP, PLM, RJA, JCC. Methodology: AKP, PLM, TAA, GCR, WJR, RJA, JCC. Software: AKP, TAA, PLM. Validation: PLM, TAA, RJA. Formal analysis: AKP, PLM, TAA, GCR, WJR, RJA, JCC. Investigation: AKP, PLM, TAA, GCR, WJR, RJA. Resources: TAA, PLM, GCR. Data curation: TAA, GCR, WJR, AKP, JCC. Writing – original draft: AKP, PLM, TAA, GCR, WJR, RJA. Writing review & editing: AKP, PLM, TAA, GCR, WJR, RJA, JCC. Visualization: PLM, TAA, GCR. Supervision: AKP, PLM, TAA, GCR. Project administration: TAA, GCR, AKP, RJA. Funding acquisition: Not applicable. Funding the authors received no funding for this work. Data availability the datasets generated and/or analyzed during the current study are openly available in the Dataverse repository at https://doi.org/10.7910/DVN/CWQA3G. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement the datasets generated and/or analyzed during the current study are openly available in the Dataverse repository at https://doi.org/10.7910/DVN/CWQA3G.