Projected Urban Air Pollution in Riyadh Using CMIP6 and Bayesian Modeling

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Comments and Suggestions:
1. While the introduction provides a general overview of air pollution challenges, it lacks a detailed discussion of specific regional factors affecting Riyadh’s air quality. Expand the discussion on Riyadh’s unique environmental and socioeconomic conditions, including its desert climate, urban sprawl, and heavy reliance on fossil fuels.
2. Add more recent and region-specific references to enhance the relevance of the background.
3. Clarify how this study fills gaps left by previous research in Riyadh and similar arid urban areas.
4. The choice of SSP scenarios (SSP2-4.5 and SSP5-8.5) is justified, but there is no explanation for why intermediate scenarios (e.g., SSP1-2.6) were excluded. Including these could provide a broader perspective on potential outcomes.
5. Discuss the limitations of relying solely on modeled projections without incorporating real-time validation data beyond 2015.
6. The study focuses primarily on SO₂ and NO₂ concentrations, with limited attention to other critical pollutants. Extend the analysis to include PM2.5 and O₃ concentrations, as these pollutants are highly relevant to human health and climate interactions.
7. Provide a step-by-step explanation of the bias-correction techniques applied to the climate models.
8. Include a sensitivity analysis to assess how changes in prior distributions or model parameters affect the results.
9. Expand the description of the data preprocessing steps, specifying thresholds for outlier removal and interpolation methods used.
10. Some figures (e.g., Figures 1–6) lack sufficient annotations or legends, making them difficult to interpret without referring back to the text.
11. The results focus heavily on high-emission scenarios (SSP5-8.5), with less emphasis on intermediate scenarios (SSP2-4.5).
12. The conclusions do not fully address the uncertainties highlighted in the results (e.g., wide credible intervals, inter-model variability).

Author Response

Comment N°1 :
While the introduction provides a general overview of air pollution challenges, it lacks a detailed discussion of specific regional factors affecting Riyadh’s air quality. Expand the discussion on Riyadh’s unique environmental and socioeconomic conditions, including its desert climate, urban sprawl, and heavy reliance on fossil fuels.

Response to Comment 1:
We thank you for this relevant and constructive comment. In accordance with your suggestion, we have substantially revised the introduction by adding a detailed discussion of the environmental and socioeconomic specificities of Riyadh. (Please refer to the newly added paragraph in blue on page 2 of the manuscript.)

«While air pollution is a global concern, Riyadh exhibits distinct environmental and socioeconomic characteristics that exacerbate challenges related to air quality. The city is subject to an extreme desert climate, marked by minimal annual precipitation and high temperatures—conditions that favor frequent sandstorms and the responsiveness of fine particulate matter. These natural phenomena significantly contribute to elevated PM2.5 concentrations, often surpassing the guidelines established by the World Health Organization (WHO) [8,9]. In parallel, Riyadh is experiencing rapid population growth and urban expansion, resulting in pronounced urban sprawl and a strong dependence on private vehicles. This urban structure, combined with a limited public transportation system, leads to substantial emissions of traffic-related air pollutants [10]. Furthermore, the city’s economy is heavily reliant on fossil fuels, both for electricity generation and industrial operations, which contributes to significant emissions of SO2, NO2, and other harmful pollutants [11,12]. Major construction activities associated with urban modernization and infrastructure projects under the Vision 2030 initiative also represent considerable sources of particulate matter [13,14]. However, the limited and fragmented air quality monitoring infrastructure in Riyadh complicates accurate exposure assessments and impedes the development of effective mitigation strategies [15]. Collectively, these factors position Riyadh as a unique case study, where the interaction between natural conditions, urban development, and energy dependence substantially heightens both environmental and public health risks associated with air pollution.»

Comment 2:
Add more recent and region-specific references to enhance the relevance of the background.

Response to comment 2:
Thank you for this valuable comment. In response, we have added more recent and region-specific references, particularly focusing on arid environments and the Riyadh region, to enhance the contextual relevance and strengthen the background section.

 

Comment N°3:

Clarify how this study fills gaps left by previous research in Riyadh and similar arid urban areas.

Response to Comment 3:

Thank you for your insightful comment. In response, we have added a dedicated paragraph on page 2 of the Introduction section (highlighted in red in the revised manuscript) to explicitly clarify how this study addresses the existing gaps in the literature.

The added paragraph reads as follows:

“Previous research on air pollution in Riyadh and comparable arid urban environments has primarily focused on the spatiotemporal characterization of fine particulate matter (PM₂.₅ and PM₁₀) and the analysis of specific episodes related to dust storms or local emissions [8, 9]. It has rarely addressed future interactions between climatic and socio-economic dynamics. Moreover, air quality monitoring remains fragmented and unsystematic, hindering a detailed understanding of local variations and exposure levels [10]. The use of Shared Socioeconomic Pathways (SSPs), despite being widely recognized as a framework for exploring future climate and socio-economic scenarios, remains limited in this regional context [1].

To address these gaps, this study adopts an integrated approach combining recent air pollution data with climate projections derived from two CMIP6 models under the SSP2-4.5 and SSP5-8.5 scenarios. By employing a multi-scale Bayesian modeling framework, this methodology enables the generation of probabilistic projections of future pollutant concentrations, thereby contributing to a more robust and forward-looking assessment of environmental risks in the context of global change.”

Comment N°4:

The choice of SSP scenarios (SSP2-4.5 and SSP5-8.5) is justified, but there is no explanation for why intermediate scenarios (e.g., SSP1-2.6) were excluded. Including these could provide a broader perspective on potential outcomes

Response to Comment 4:

Thank you for your insightful comment. In the revised version of the manuscript (page 3, Materials and Methods section, text highlighted in blue), we have expanded our justification regarding the exclusive selection of the SSP2-4.5 and SSP5-8.5 scenarios. The following paragraph was added:

“The SSP2-4.5 and SSP5-8.5 scenarios were specifically selected in this study due to their strong relevance to the socio-economic and energy context of Saudi Arabia, particularly in Riyadh. SSP2-4.5 reflects an intermediate pathway based on current trends, while SSP5-8.5 represents an intensive development trajectory centered on fossil fuels. This methodological approach aligns with several recent studies conducted in the region, such as Alghamdi et al. (2022) [27], who projected PM2.5 levels in Riyadh under these two scenarios based on urban expansion and emission trends; Zhang et al. (2021) [28], who modeled temperature and air pollution extremes in Middle Eastern cities using SSP2-4.5 and SSP5-8.5; and Mahmoud et al. (2023) [29], who assessed ozone and fine particulate pollution in Gulf cities based on the same scenarios, selected for their consistency with regional energy policies. Furthermore, the marked contrast between these two scenarios — one moderate, the other extreme — allows for the coverage of a wide range of potential impacts, thereby offering a more nuanced and useful perspective for planning adaptation measures in arid environments.”

Comment N°5:

Discuss the limitations of relying solely on modeled projections without incorporating real-time validation data beyond 2015

Response to Comment 5:

Thank you for your comment. In response, we have addressed this issue by adding a dedicated paragraph (highlighted in red on page 10 of the revised manuscript, within the Discussion section). The newly added paragraph reads as follows:

“The absence of recent datasets, specifically those after 2015, limits the ability to verify and adjust model outputs based on actual atmospheric conditions observed in Riyadh, particularly in a context of rapid urbanization and evolving emission sources [67]. To mitigate this limitation, rigorous methods for correcting systemic biases were applied, in accordance with best practices in climate modeling [75, 76]. These corrections adjust model outputs by accounting for discrepancies identified between historical simulations and available observations (2000–2015), thereby reducing uncertainties related to physico-chemical parameterizations and anthropogenic forcings. The probabilistic approach employed, which includes hierarchical credibility intervals, allows for the explicit quantification of residual uncertainty while preserving the robustness of projected trends. Although the lack of recent validation data may limit fine-tuning of the models, the bias correction procedures ensure that the projections remain consistent with the dominant atmospheric mechanisms observed in arid environments [74]. This methodology, combined with inter-model agreement on the significant degradation of air quality, strengthens the reliability of the conclusions for the development of adaptation policies.”

Comment6
The study focuses primarily on SO₂ and NO₂ concentrations, with limited attention to other critical pollutants. Extend the analysis to include PM2.5 and O₃ concentrations, as these pollutants are highly relevant to human health and climate interactions.

Response to Comment 6
We thank you for this relevant observation. In fact, future concentrations of PM₂.₅ and O₃ were analyzed in Section 3.5, entitled “Spatial and temporal variability of PM₂.₅ and O₃ pollution in future emission scenarios” (pages 13–18). This section provides a detailed analysis of the spatial and temporal aspects of projected pollution for these two pollutants.

In addition, a new paragraph has been added on pages 19–20 in the Discussion section to examine and interpret the results related to future PM₂.₅ and O₃ concentrations.

Please find attached:

• The part added to the “Results” section(Section 3.5)
• The part added to the “Discussion” section

Results section

3.6. Spatial and temporal variability of PM2.5 and O3 pollution in future emission scenarios

Climate projections for the period 2030–2070 reveal a marked increase in tropospheric ozone (O3) and fine particulate matter (PM2.5) concentrations across the Riyadh region   [27 ]. This trend, particularly pronounced under the high-emission scenario SSP5-8.5 [70 ], was observed in both the CNRM [ 71 ] and MPI [ 72 ] climate models. This deterioration results from the combined effects of climate change, intensified anthropogenic activities, and rapid urban expansion [29].

3.6.1. PM2.5 Concentration analysis

A comparison between the historical and projected PM2.5 concentrations revealed significant temporal and spatial variations across the study domain (Figure 7). Historical data (2000–2014) from both CNRM-ESM2-1 and MPI-ESM1.2 models show relatively uniform concentrations across regions R1–R6, with values ranging from 35–50 μg/m3. The CNRM-ESM2-1 model indicates slightly higher historical concentrations (40–48 μg/m3) ?compared to MPI-ESM1.2 (35–45 μg/m3), with R3 showing the highest historical values in both models

 

Figure 7. Regional PM2.5 concentrations simulated by CMIP6 models (CNRM-ESM2-1 and MPI-ESM1.2) for the historical period (2000–2015) and future projections (2030–2070) under SSP5-4.5 and SSP5-8.5 scenarios.

The comprehensive statistical analysis of PM2.5 concentrations across all regional models reveals distinct patterns between climate scenarios and models (Table 4). Under the SSP2-4.5 moderate emission scenario (2030–2070), both models projected substantial ncreases, albeit with notable differences. The CNRM-ESM2-1 model projects concentrations of 50–62 μg/m3 across regions, with R3 reaching the highest values (∼62 μg/m3), followed by R1 and R5 (∼58 μg/m3). In contrast, the MPI-ESM1.2 model showed more conservative ?projections (48–58 μg/m3), with R3 again showing peak concentrations. Regions R2, R4, and R6 consistently showed lower concentrations in both models under this scenario.

Table 4. Estimated mean PM2.5 concentrations (μscenario) and standard deviations (SD) under historical and future climate scenarios from two climate models.

Scenario (Model)	Mean (μscenario)	SD	Min	Max
Historical – CNRM-ESM2-1	45.13	4.25	38.9	52.6
Historical – MPI-ESM1.2	42.53	4.89	36.4	49.8
SSP2-4.5 – CNRM-ESM2-1	58.90	5.68	51.2	67.4
SSP2-4.5 – MPI-ESM1.2	55.97	5.37	48.7	64.1
SSP5-8.5 – CNRM-ESM2-1	73.33	7.14	64.8	83.7
SSP5-8.5 – MPI-ESM1.2	69.97	6.63	61.9	79.9

 

The SSP5-8.5 high-emission scenario revealed the most severe deterioration, with marked differences between models and regions. The CNRM-ESM2-1 model projects concentrations reaching 60–68 μg/m3, with R3 and R5 showing the highest values (∼68 μg/m3 and ∼65 μg/m3 respectively). The MPI-ESM1.2 model projects slightly lower but still critical concentrations (58–65 μg/m3), maintaining R3 as the most polluted region (∼65 μg/m3).

When compared to historical data, this represents an increase of 35–70% depending on the region and model, as evidenced by the mean concentrations presented in Table 4.

3.6.2. O3 Concentration projections

Ozone concentrations showed even more dramatic increases across all scenarios and models (Figure 8). Historical concentrations (2000–2015) were relatively uniform across regions R1–R6, with both models indicating concentrations between 85 and 95 μg/m3. The CNRM-ESM2-1 model shows slightly higher historical values (88–95 μg/m3) compared to MPI-ESM1.2 (85–92 μg/m3), with minimal spatial variation. The statistical summary of O3 concentrations demonstrates the severity of future ozone pollution across all scenarios (Table 5). Under the SSP2-4.5 scenario, significant increases were observed with clear spatial differentiation. The CNRM-ESM2-1 model projects concentrations of 95–115 μg/m3, with R3 and R5 showing the highest values (∼115 μg/m3 and ∼112 μg/m3 respectively). The MPI-ESM1.2 model indicated similarspatial patterns but with slightly lower absolute values (92–110 μg/m3). Regions R1, R2, R4, and R6 maintained more moderate concentrations (95–105 μg/m3) in both models.

Table 5. Estimated mean O3 concentrations (μscenario) and standard deviations (SD) under historical and future climate scenarios from two climate models

Scenario (Model)	Mean (μscenario)	SD	Min	Max
Historical – CNRM-ESM2-1	95.33	5.65	88.4	103.2
Historical – MPI-ESM1.2	91.98	5.50	85.2	99.6
SSP2-4.5 – CNRM-ESM2-1	118.53	6.92	109.7	128.4
SSP2-4.5 – MPI-ESM1.2	114.22	6.63	105.8	123.7
SSP5-8.5 – CNRM-ESM2-1	142.02	8.64	131.2	153.9
SSP5-8.5 – MPI-ESM1.2	136.78	8.21	126.4	148.2

The SSP5-8.5 scenario projected critical ozone levels across all regions. The CNRM-ESM2-1 model indicates concentrations reaching 125–142 μg/m3, with R3 showing the most severe pollution (∼142 μg/m3), followed closely by R5 (∼138 μg/m3). The MPI-ESM1.2

 

Figure 8. Regional O3 concentrations simulated by CMIP6 models (CNRM-ESM2-1 and MPI ESM1.2) for the historical period (2000–2015) and future projections (2030–2070) under SSP5-4.5 and SSP5-8.5 scenarios

model projected similarly alarming levels (120–138 μg/m3), maintaining the same spatial pattern. Compared to historical data, these projections represent increases of 40–60% across all regions, as summarized in the statistical analysis presented in Table 5.

3.6.3. Model-scenario-regional comparisons

The comparative analysis revealed consistent patterns across both climate models, despite the quantitative differences. The CNRM-ESM2-1 model systematically projects higher concentrations for both pollutants compared to MPI-ESM1.2, with differences ranging from 2–8 μg/m3 for PM2.5 and 3–10 μg/m3 for O3. However, both models agree on the spatial ranking of regions, with R3 consistently emerging as the most polluted region under all scenarios.

The scenario comparison shows that SSP5-8.5 projections exceed SSP2-4.5 values by 8–15 μg/m3 for PM2.5 and 15–25 μg/m3 for O3 across all regions and models. This demonstrates the critical importance of emission pathway selection for future air quality outcomes, as clearly illustrated by the mean concentration differences between scenarios in Tables 4 and 5. Regional analysis revealed that R3 and R5 consistently showed the highest pollution levels under future scenarios, whereas R2, R4, and R6 maintained relatively lower concentrations. R1 exhibited intermediate pollution levels across all scenarios. This spatial pattern suggests that urban development and industrial activities will concentrate in specific areas, creating distinct pollution gradients across metropolitan regions.

3.6.4. Health and environmental implications

These projected concentrations have severe health implications when compared with the WHO guidelines [ 73]. For PM2.5, future concentrations will exceed the WHO annual guideline (5 μg/m3) by factors of 10–14 under SSP2-4.5 and 12–16 under SSP5-8.5. Even the WHO interim target (15 μg/m3) will be surpassed by factors of 3–4.5 across all regions and scenarios. For ozone, the projected concentrations substantially exceeded the WHO 8-hour guideline (100 μg/m3) under both scenarios, with exceedances of 10–42 μg/m3 depending on the region and scenario. The WHO annual guideline for ozone (60 μg/m3) will be exceeded by factors of 1.6–2.4 across all projections, indicating widespread health risks for the population.

3.6.5. Temporal evolution and regional hotspots

The temporal evolution analysis revealed distinct patterns of pollution development across different regional clusters (Figures 10 and 9). The integration of temporal data from the comprehensive time-series analysis provides critical insights into the progression of air pollution across the Riyadh metropolitan region.

 

Figure 9. Temporal evolution of PM2.5 concentrations (2030-2070) across different regional clusters in the Riyadh region under SSP2-4.5 and SSP5-8.5 scenarios using CNRM-ESM2-1 and MPI-ESM1.2 climate models.

PM2.5 Temporal trajectories

The temporal analysis of PM2.5 concentrations across different regional clusters reveals alarming upward trends with distinct patterns for each geographical zone. The highly urbanized core regions (R11, R14, R15) demonstrate the most critical pollution trajectory, with concentrations projected to increase from baseline levels of 45–50 μg/m3 in 2030 to 90–130 μg/m3 by 2070 under SSP5-8.5. This represents a doubling of concentrations over the 40-year period, with a consistent annual increase rate of 1.5–2 μg/m3 per year.

The industrial and commercial zones (R12, R13) exhibit similarly concerning trends, with concentrations rising from initial values of 40–45 μg/m3 in 2030 to 85–110 μg/m3 by 2070. These regions show a linear growth pattern with noticeable acceleration after 2050, suggesting the compounding effects of industrial expansion and climate change impacts.

The central urban area (R10) displays intermediate pollution levels, with concentrations increasing from 35–40 μg/m3 in 2030 to 75–90 μg/m3 by 2070. This region maintains a steady but moderate growth rate throughout the projection period. Peripheral regions show varied patterns: R1 and R4 maintain the lowest absolute concentrations but still experience significant increases from 30–35 μg/m3 to 65–85 μg/m3.

Regions R2, R3, and R5–R9 follow intermediate trajectories, with concentrations rising from 20–30 μg/m3 in 2030 to 45–70 μg/m3 by 2070.

 

Figure 10. Temporal evolution of O3 concentrations (2030-2070) across different regional clusters in the Riyadh region under SSP2-4.5 and SSP5-8.5 scenarios using CNRM-ESM2-1 and MPI-ESM1.2 climate models.

O3 Temporal evolution

The ozone concentration projections reveal even more dramatic temporal changes, with critical implications for public health. The highly urbanized core regions (R11, R14, R15) emerge as extreme pollution hotspots, with concentrations projected to increase from baseline levels of 150–160 μg/m3 in 2030 to 350–450 μg/m3 by 2070 under SSP5-8.5. This represents a 2.5 to 3-fold increase, with the steepest temporal gradients showing annual increases of 5–7 μg/m3 throughout the projection period. The industrial and commercial zones (R12, R13) display similarly alarming trends, with ozone concentrations rising from 100–110 μg/m3 in 2030 to 250–350 μg/m3 by 2070.

Notably, the MPI-ESM1.2 model consistently projects higher concentrations in these regions compared to CNRM-ESM2-1, with differences reaching 20–30 μg/m3 by 2070.

The central urban area (R10) shows intermediate but still critical pollution levels, with concentrations increasing from 85–95 μg/m3 in 2030 to 200–250 μg/m3 by 2070. This region maintains a consistent upward trajectory throughout the projection period.

Peripheral regions (R1, R4, R2, R3, and R5–R9) demonstrate more moderate but still significant increases. R1 and R4 show concentrations rising from 80–90 μg/m3 to 180– 220 μg/m3, while R2, R3 increase from 60–75 μg/m3 to 140–180 μg/m3, and R5–R9 rise from 70–85 μg/m3 to 160–200 μg/m3 by 2070.

Inter-Scenario temporal comparison

The comparative analysis between SSP2-4.5 and SSP5-8.5 scenarios reveals increasingly divergent trajectories over time. For PM2.5, the difference between scenarios grows from 10–15 μg/m3 in 2030 to 20–30 μg/m3 by 2070, representing a 25–35% higher concentration under the high-emission scenario. This divergence becomes more pronounced in the latter half of the projection period, emphasizing the long-term consequences of emission pathway choices.

For ozone, the scenario differences are even more dramatic, with gaps widening from 15–25 μg/m3 in 2030 to 40–60 μg/m3 by 2070. The SSP5-8.5 scenario consistently produces concentrations 30–40% higher than SSP2-4.5 across all regional clusters.

Model Agreement and Uncertainty

The temporal analysis confirms that both CNRM-ESM2-1 and MPI-ESM1.2 models consistently identify the same regional hierarchy in terms of pollution severity, with the highly urbanized core areas (R11, R14, and R15) representing the most critical pollution hotspots. However, notable inter-model differences emerge in the temporal trajectories: the MPI model tends to project 10–20% higher concentrations for ozone, particularly in industrial zones, while CNRM shows slightly higher PM2.5 projections in urban areas.

Despite these quantitative differences, the qualitative agreement between models reinforces the robustness of the overall trends and the identification of critical pollution hotspots. Both models consistently show accelerating pollution trends after 2050, suggesting that the combined effects of climate change and urban development will intensify air quality deterioration in the latter half of the projection period.

Discussion  section

The results indicate a pronounced and spatially uneven degradation of air quality in Riyadh between 2030 and 2070, especially under SSP5-8.5. Historical PM2.5 levels already exceeded WHO guidelines by a factor of 7–10, and are projected to rise by an additional 5–70%, with concentrations reaching 70–83 μg/m3 in urban and industrial zones such as R3, R5, R11, R14, and R15. Ozone (O3) is expected to follow a similar trajectory, exceeding 140 μg/m3 under SSP5-8.5—40–60% above current levels and well beyond the WHO 8-hour guideline. The divergence between SSP2-4.5 and SSP5-8.5 becomes particularly stark after 2050. CNRM-ESM2-1 and MPI-ESM1.2 models show consistent spatial and temporal patterns despite a 10–20% difference in magnitude, with CNRM projecting higher PM2.5 and MPI higher ozone. These levels pose serious health risks, particularly to vulnerable populations, and call for targeted mitigation, air quality monitoring, and integrated air–climate strategies

Comment N°7:

Provide a step-by-step explanation of the bias-correction techniques applied to the climate models.

Response to Comment 7:

Thank you for your comment. In response, we have added an explanatory paragraph in the Materials and Methods section, highlighted in blue on page 5. This paragraph details the linear bias correction method applied to the simulated daily concentrations of NO₂, SO₂, and PM₂.₅. Please find the added text below:

«In parallel, a linear bias correction was applied to the daily simulated concentrations of atmospheric pollutants (NO₂, SO₂, and PM₂.₅) to ensure consistency with local observations in Riyadh. For each pollutant and monitoring station, the average daily bias Δ over the historical period (2000–2015) was calculated as:

Δ = C̄_obs - C̄_sim                                  (1)

where C̄_obs and C̄_sim represent the mean observed and simulated concentrations, respectively.
This bias value was then used to correct future daily simulated concentrations according to:

C_corr(t) = C_sim(t) + Δ                      (2)

where C_sim(t) is the uncorrected simulated concentration at time t, and C_corr(t) is the bias-corrected value.
This additive correction method effectively reduces systematic mean bias while preserving daily variability. Cross-validation against historical data showed a marked improvement in the agreement between corrected simulations and observations. Therefore, this approach improves the robustness of projected daily concentrations of NO₂, SO₂, and PM₂.₅ for the Riyadh urban environment [36–39].»

Comment 8:

Include a sensitivity analysis to assess how changes in prior distributions or model parameters affect the results.

Response  to comment 8

Thank you very much for this comment. In response, we have incorporated a detailed sensitivity analysis to assess the impact of variations in prior distributions and model parameters on our results.

Two paragraphs have been added: the first, in brown color, appears in the Materials and Methods section under the title Sensitivity analysis, where we describe the methodological approach adopted for this analysis; the second, also in brown, has been inserted in the Results section under the title Sensitivity analysis to assess the impact of prior distributions and model parameters, presenting the main findings of this analysis. You will find these two added paragraphs attached.

2.4. Sensitivity analysis

To evaluate the robustness of the Bayesian hierarchical model and ensure the reliability of future air pollution projections in Riyadh, a comprehensive sensitivity analysis was performed. This analysis assessed the impact of various methodological choices, including alternative prior distributions for scenario-specific mean effects (μscenario), different likelihood specifications (normal, log-normal, gamma, and Student-t distributions), assumptions regarding variance structure (homoscedastic versus heteroscedastic), and multiple strategies for integrating outputs from the CNRM-ESM2-1 and MPI-ESM1.2 climate models (equal weighting, performance-based weighting, and hierarchical modeling). Additionally, the numerical stability of the MCMC sampling process was evaluated by varying chain length, burn-in duration, and initialization methods. Sensitivity was assessed using three key evaluation criteria: (i) relative changes in posterior means exceeding 10%, (ii) credible interval overlap below 80%, and (iii) stability in the ranking of scenarios or spatial units.

Particular attention was given to the consistent identification of high-risk regions (R3,

R5, R11, R14, R15) and to the robustness of exceedance probabilities for WHO air quality guidelines under alternative modeling choices. This sensitivity analysis ensures that key findings are not artifacts of specific prior, likelihood, or integration assumptions, thereby enhancing the credibility of the projections and their utility for informing environmental policy in arid urban environments.

3. Results

3.1. Sensitivity analysis to assess the impact of prior distributions and model parameters

The sensitivity analysis confirmed the robustness of the conclusions derived from the Bayesian hierarchical model across a wide range of methodological choices. Changes in prior distributions resulted in less than 8% variation in the posterior means of scenario effects, while alternative specifications of global standard deviation hyperparameters in duced limited discrepancies, remaining below 15%. Alternative likelihood distributions (log-normal, gamma, and Student-t) provided improved model fit for PM2.5 and O3, moderately widening the credible intervals (up to +10%) without affecting central trends. For NO2 and SO2, model outputs remained stable across all tested likelihood forms, underscoring their resilience to distributional assumptions. The introduction of heteroscedastic variance assumptions revealed a 20–30% increase in uncertainty under the SSP5-8.5 scenario, particularly for PM2.5 and O3, reflecting increased sensitivity to high-emission trajectories.

Comment 9:

Expand the description of the data preprocessing steps, specifying thresholds for outlier removal and interpolation methods used

We thank you for this insightful comment. In response, a paragraph has been added in green within the Materials and Methods section, detailing the data preprocessing steps. This paragraph specifically explains the criteria used to identify and remove outliers based on the 3IQR rule, as well as the interpolation methods applied to missing values—distinguishing between short gaps (up to three consecutive days) and longer gaps (more than three days). The newly added paragraph is included in the revised manuscript.

Outlier values in the daily time series for each pollutant were identified and removed using the 3IQR rule: any daily value below Q1 − 3 × IQR or above Q3 + 3 × IQR, where Q1 andQ3 represent the 25th and 75th percentiles respectively, and IQR = Q3 − Q1, was excluded from the analysis. For missing daily values, linear interpolation was applied for short gaps (up to three consecutive days). For longer gaps, missing values were imputed using the mean of the same calendar day over the two adjacent weeks (±2 weeks), to preserve seasonal and weekly patterns. All analyses were conducted using these quality-controlled daily average concentrations, ensuring temporal consistency across all pollutants and monitoring station

Comment10:
Some figures (e.g., Figures 1 to 6) lack sufficient annotations or legends, making them difficult to interpret without referring to the text.

Response to comment 10:
Thank you for this remark. In response, we have revised Figures 1 to 6 by adding clearer annotations and more detailed legends.

Comment 11:

The results focus heavily on high-emission scenarios (SSP5-8.5), with less emphasis on intermediate scenarios (SSP2-4.5).

Response to comment 11:

We thank you for this remark. In the revised version of the Results and Discussion section, we have carefully incorporated a detailed and comparative analysis of the intermediate (SSP5-4.5) and high (SSP5-8.5) scenarios. These additions, including numerical data, tables, figures, and summaries, help to better highlight the differences in impact on SO2 and NO2 concentrations, as well as the benefits of intermediate pathways in terms of reducing environmental risks and supporting policy decision-making.

Comment 12:

The conclusions do not fully address the uncertainties highlighted in the results (e.g., wide credible intervals, inter-model variability).

Response 12

Thank you for this relevant observation. In response, we have revised the conclusion to explicitly address the uncertainties raised in the results. A dedicated paragraph, highlighted in blue in the revised version of the manuscript, has been added. Please find below the inserted paragraph:

The 95% Bayesian credible intervals for the projected mean SO2 concentrations under the SSP5-8.5 scenario ranged from –0.37 to +0.60ppb for the CNRM-ESM2-1 model and from  0.35 to +0.53ppb for the MPI-ESM1.2 model. For NO2, the intervals extended from –0.58 to +0.54ppb, while inter-model differences reached up to 0.16ppb for SO2 and 0.02ppb for NO2, highlighting the structural uncertainty inherent in the diversity of climate models used. Sensitivity analyses indicated that variations in the prior distributions and model parameters resulted in less than an 8% change in the posterior means. Despite these uncertainties, the identification of high-risk zones (R3, R5, R11, R14, and R15) remained robust in over 95% of cases, and the high probabilities of exceeding WHO thresholds (ranging from 80% to 98%) reinforced the reliability of the observed trends. Overall, the general trends and spatial distribution of risks appear statistically sound and offer a reliable foundation for informing public policies and developing targeted mitigation strategies.

Author Response File: Author Response.doc

Reviewer 2 Report

Comments and Suggestions for Authors

Attached my comments related with the manuscript. Authors should pay attention because a lot of mistakes related with the pollutants analyzed exist in the document (NO-NO2, SO-SO2). Furthermore, the methodology and data used should be justified. Why this methodology, why these models...Performance, accuracy, uncertainty are terms that may be presents in the study and authors are invited to add. Thanks.

Comments for author File: Comments.pdf

Author Response

Comment 1:

Why do you use NO? NO2 is one of the most relevant pollutants in urban environments. NO generated by the traffic combines with O2 generating NO2 in the cities.

Response to Comment1:

We thank you for your comment. This was indeed a typographical error on our part. We were referring to NO₂, which, as you rightly pointed out, is a major pollutant in urban environments. The mentions of NO and SO have been corrected and replaced by NO₂ and SO₂, respectively, throughout the revised version of the manuscript.

Comment2 :
Several exclamation marks appear in the text in place of the reference numbers, indicating a possible LaTeX compilation issue.

Response to the comment 2:
The exclamation marks that appeared in the manuscript were due to a LaTeX compilation issue related to the bibliographic references. In the revised version, this problem has been corrected, and the reference numbers now appear properly in the text.

Comment3 :

who public administration manage these stations? Please to add.

Response to the comment 3:

We thank you for your comment. To answer your question regarding the management of the stations, we have added the following sentence in red in the manuscript:

"The air quality monitoring stations are managed by the Royal Commission for the City of Riyadh."

Comment4 :

Indicate North/South, East /West in the table

Response to the comment 4:

We thank you for this pertinent comment. In response, we have added the cardinal direction indicators North (N) and East (E) to the latitude and longitude values in the monitoring stations table (Table 1 in the revised version).

 

 

Monitoring stations in Riyadh with merged zone type names for similar zones.

Code	Area Name	Zone Type	Latitude (N)	Longitude (E)
R1	National Guard Hospital	Healthcare Facility	24.72° N	46.71° E
R2	Al Hair	Semi-natural Area	24.53° N	46.66° E
R3	Wadi Hanifa	Semi-natural Area	24.58° N	46.60° E
R4	Al Aziziyah Hospital	Healthcare Facility	24.59° N	46.80° E
R5	Al Shifa	Residential District	24.57° N	46.81° E
R6	Al Zamroud	Residential District	24.60° N	46.70° E
R7	Al Amal	Residential District	24.71° N	46.72° E
R8	Al Zara	Residential District	24.71° N	46.67° E
R9	Al Mauroj	Residential District	24.76° N	46.71° E
R10	King Faisal	Urban Area	24.69° N	46.71° E
R11	Eastern Ring Road	Traffic Corridor	24.69° N	46.80° E
R12	King Fahad Road	Dense Urban Area	24.72° N	46.65° E
R13	Makkah Road	Dense Urban Area	24.64° N	46.60° E
R14	Northern Ring Road	Traffic Corridor	24.80° N	46.70° E
R15	Southern Ring Road	Traffic Corridor	24.56° N	46.85° E

Comment 5:

 why these two models were selected? Please to justify.

Response to the comment5:

We thank you for your comment. To clarify and justify the choice of climate models used in our study, we have added a paragraph detailing the scientific strengths of the CNRM-ESM2-1 and MPI-ESM1.2 models. This paragraph has been added in red in Section 2.2: Climate projection models and atmospheric data processing.

These models are among the latest generation of Earth System Models contributing to CMIP6, widely used in climate impact and air quality studies due to their comprehensive representation of atmospheric chemistry and climate processes. Both models have been evaluated for their performance in simulating regional climate patterns and pollutant dynamics, making them suitable for assessing future air quality under different emission scenarios. For example, Voldoire et al. (2019) present the development and evaluation of CNRM-ESM2-1, highlighting its improved atmospheric chemistry module and ability to simulate aerosol and trace gas concentrations relevant for air quality assessments [35]. Similarly, Giorgetta et al. (2018) describe the MPI-ESM1.2 model’s enhanced climate sensitivity and chemistry-climate interactions, which are critical for robust projections of pollutants under changing climate conditions [ 36]. Moreover, multi-model approaches including these models have been recommended to capture structural uncertainties inherent in climate projections [37]. Using both models thus strengthens the robustness of the study’s conclusions by encompassing a range of plausible climate-chemistry responses.

 

Comment 6

Please to indicate previously when AQ stations are presented

Response to the comment 6:

Thank you for your comment. We have included a detailed description of the air quality monitoring network in Riyadh in the manuscript. The following paragraph has been added in the Materials and Methods section, prior to the presentation of the data:

The air quality monitoring network in Riyadh consists of multiple stations strategically located across different urban zones, including residential, traffic corridors, healthcare facilities, and semi-natural areas. Table 1 summarizes the main monitoring stations used in this study, detailing their geographic coordinates and zone types. These stations provide continuous measurements of key pollutants such as PM10, PM2.5, NO2, and O3, enabling a comprehensive spatial and temporal assessment of air quality in Riyadh.

Comment 7

which is the uncertainty of the measures used? Authors should include information about this.

We thank you for your comment. In response, we have inserted a detailed explanation of the air quality data processing and uncertainty estimation in the appropriate section of the manuscript. The added text appears in blue in the revised version.

The uncertainty of the air quality measurements used in this study was assessed according to the ISO 20988 and ISO 11222 standards. The typical expanded uncertainties (at the 95% confidence level) were estimated at approximately 10–15% for PM, 10% for NO2, and 10–12% for O3, thereby ensuring the reliability of the data used for spatial and temporal analyses

Comment 8

Authors should to explain the different methods that exists to analyse the AQ in a future climate context. Tipically, are used photochemical models coupled with meteorological models that use the information provided by climate change global models. Doing in this case a dynamical downscaling. Authors have been selected other methodology and should explain the advantages and disadvantages, limitations, benefits, etc of everyone of them.

Response to the comment 8:

We thank the reviewer for this insightful comment. In response, we have added the following paragraph to the manuscript to explain the rationale behind the selected methodology and to highlight its advantages and limitations compared to traditional dynamical downscaling approaches:

In this study, we adopted a hybrid approach combining the statistical bias correction of climate models and hierarchical Bayesian modeling to project future atmospheric pollutant concentrations in Riyadh. Outputs from the global climate models CMIP6 (CNRM-ESM2-1 and MPI-ESM1-2-HR) were first statistically corrected using linear adjustment techniques to align the simulations with local observations and improve the regional relevance of the projections [ 38 ,39 ]. An additive correction was then applied to the simulated concentrations for each pollutant and monitoring station. A hierarchical Bayesian model explicitly integrates uncertainties related to climate models, regional spatial variability, and interannual dynamics, thereby providing robust credible intervals for each region/scenario combination [45 –47]. This methodology has the advantage of being less computationally demanding than traditional dynamical downscaling while allowing explicit quantification of uncertainties and better adaptation to local data.

Comment 9

Reliability of climate model outputs: The climate models used (CNRM-ESM2-1 and MPI-ESM1-2-HR) provide detailed and validated outputs primarily for this period, ensuring robust climate and air quality analyses.

Response to the comment 9:

We thank you for this relevant observation regarding the duration of the analysis period. Indeed, we used a shorter period than the recommended 30 years due to several methodological and practical constraints:

Data availability: High-resolution and homogeneous-quality data covering all urban areas studied are only available for the 2000–2015 period.

Temporal relevance: This recent period is particularly representative of current urban dynamics, which are characterized by rapid growth, infrastructure transformation, and increasing environmental pressures.

Reliability of climate model outputs: The climate models used (CNRM-ESM2-1 and MPI-ESM1-2-HR) provide detailed and validated outputs primarily for this period, ensuring robust climate and air quality analyses.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

I am pleased to see the significant improvements in this revised version of your manuscript. The authors effectively addressed all the concerns raised in my previous round of reviews, and the clarity and presentation have been greatly enhanced, making the paper much stronger.

 

Reviewer 2 Report

Comments and Suggestions for Authors

Very appreciated the changes done. I think that the research now is more clear.