Assessment of PM10 and PM2.5 Concentrations in Santo Domingo: A Comparative Study Between 2019 and 2022

1. Introduction

Air pollution remains one of the most pressing environmental and public health challenges in urban areas worldwide [1,2,3,4,5,6,7,8,9]. Particulate matter (PM) consists of tiny solid particles and liquid aerosols containing acids, organic compounds, metals, and dust [10,11,12,13,14]. PM originates from both natural sources, such as volcanic eruptions, wildfires, dust storms, and Saharan air layer, and human activities like fuel combustion, vehicle exhaust, and industrial emissions [11,14,15,16,17,18,19,20,21].

PM, particularly PM₁₀ (particles with a diameter less than 10 $\mu m$) and PM_2.5 (particles with a diameter less than 2.5 $\mu m$), is of special concern due to its adverse effects on human health and its ability to penetrate the respiratory system [22,23,24]. Elevated concentrations of PM have been associated with increased morbidity and mortality from cardiovascular and respiratory diseases, making it a significant topic of study in environmental science [25,26,27,28,29,30].

Urban environments, especially in developing regions, are particularly susceptible to high levels of PM due to a combination of vehicular emissions, industrial activities, construction, and the influence of meteorological conditions [7,8,31,32,33]. In this context, identifying the spatial and temporal patterns of particulate matter, as well as its key predictors, is essential to understanding the factors driving air quality in urban areas and to informing effective mitigation strategies [9,34,35,36,37,38]. Studies have shown that variables such as wind speed, air pressure, temperature, rainfall, and aerosol optical depth (AOD) can influence PM concentrations, but their relative importance often varies by location, season, and particle size [37,39,40].

The city of Santo Domingo, located in the National District on the south-central coast of the Dominican Republic, is an ideal setting for this study due to its diverse urban environments and increasing exposure to air pollution [41,42,43,44,45,46,47]. However, comprehensive studies investigating the spatial distribution of PM and its relationship with meteorological and environmental variables in this region are still limited [42]. Previous work has highlighted the need for systematic monitoring and analysis of PM concentrations, including the use of low-cost sensors, to better understand local air quality dynamics [21,42,47,48,49].

In this study, we analyze PM₁₀ and PM_2.5 concentrations collected during two distinct sampling campaigns in 2019 and 2022 across various urban environments in Santo Domingo. The primary objectives of this research are: (1) to assess the spatial distribution and temporal variability of PM₁₀ and PM_2.5, (2) to investigate the relationship between PM concentrations and key meteorological, environmental variables and satellite observations using linear regression models, and (3) to determine the predictive power of these variables for explaining PM variations in both years. By identifying significant predictors of PM concentrations, this study contributes to a better understanding of air quality dynamics in Santo Domingo and provides a foundation for targeted environmental policies and interventions aimed at improving urban air quality.

2. Materials and Methods

The study was conducted in Santo Domingo, National District, Dominican Republic (ca. 18.49°N, 69.96°W), focusing on various types of urban environments (Figure 1). The sampling sites comprised public and private schools, a university, and an urban park, selected based on a previous study [42,47]. A total of 26 sampling sites were selected in 2019 for PM₁₀ collection, with each site sampled at least three times between July and December. The sampling design was systematic, aiming for a 2 km separation between points. The final locations, including the specific schools, university, or recreational areas, were determined using the nearest neighbor method based on the ideal points initially established through software. In 2022, a new sampling campaign was conducted, incorporating additional measurements. A total of 30 sites were sampled between May and July, each sampled once.

Due to logistical and budgetary constraints, both campaigns opted to conduct measurements on a single day per site. Although this strategy limits the capture of daily variations, it allows for a comparative assessment of PM₁₀ levels at different locations within the study area, providing a representative overview of the spatial distribution of pollution. To evaluate the limitation of single-day sampling per site, statistical analyses were conducted, including linear regression, correlations with meteorological variables, and spatial interpolation techniques.

Although efforts were made to maintain the same sampling sites in both campaigns, in some cases, this was not possible due to access restrictions, lack of permits, or safety concerns for equipment installation. To ensure the continuity of the study, alternative locations were selected in representative areas within the same urban environment. As noted, the second campaign included more sites to improve spatial coverage and capture variability in air pollution across different areas of Santo Domingo. This adaptation allowed for a broader perspective on the distribution of PM₁₀ in the city, complementing the data obtained in 2019. While some sampling sites varied between 2019 and 2022, the selection followed homogeneous location criteria (urban areas with similar characteristics). Additionally, statistical analyses were applied to assess general trends in PM₁₀ and PM_2.5, ensuring the comparability of results between both periods.

For both sampling campaigns, particulate matter was collected using MiniVol^TM TAS Portable Air Samplers (AirMetrics Co., Oregon, USA) [50,51]. This device operates by drawing air through size-selective impactors, which separate PM₁₀ and PM_2.5 fractions based on their aerodynamic diameter. In 2019, only PM₁₀ was measured, as the campaign was initially designed to focus on coarse particles. In 2022, both PM₁₀ and PM_2.5 were measured by using the appropriate impactor for each fraction, providing a more comprehensive analysis of particulate matter.

The MiniVol TAS samplers were deployed at each site for 24 hours to collect representative samples of particulate matter. Both the 2019 and 2022 samples were single-day measurements, with no daily replicates collected. The samplers were calibrated and maintained according to the manufacturer’s specifications to ensure data accuracy and reliability.

2.1. Data Analysis

The collected data were analyzed using a variety of statistical methods to evaluate the concentration and distribution of particulate matter (PM₁₀ and PM_2.5) in Santo Domingo. The analysis included descriptive statistics, correlation analyses, regression modeling, and geospatial analysis.

Descriptive statistics were calculated to summarize the data, including measures of central tendency and dispersion. The analysis provided mean, median, standard deviation, and confidence intervals for PM₁₀ and PM_2.5 concentrations across different sampling periods.

Correlation analyses were performed to examine the relationships between PM₁₀ and PM_2.5 concentrations [36,52]. Pearson and Spearman correlation coefficients were calculated to assess the strength and direction of the associations. The Pearson correlation coefficient, r, is given by the equation 1

$$r={\displaystyle \frac{\sum (x_i-\overline{x})(y_i-\overline{y})}{\sqrt{\sum {(x_i-\overline{x})}^2·\sum {(y_i-\overline{y})}^2}}}$$

(1)

and the Spearman correlation coefficient, $\rho$, is given by the equation 2

$$\rho =1-{\displaystyle \frac{6\sum d_i^2}{n(n^2-1)}}$$

(2)

where $x_i$ and $y_i$ are the individual sample points, $\overline{x}$ and $\overline{y}$ are the means of the sample points, $d_i$ is the difference between the ranks of corresponding variables, and n is the number of observations.

Cross-correlation functions (CCF) were calculated to examine the temporal relationships between aerosol optical depth (AOD) and PM₁₀ concentrations, using daily data along with weekly and monthly averages for temporal aggregation [20,53,54,55,56,57,58,59]. Time series visualizations were used to represent the temporal trends in AOD and PM₁₀ concentrations for 2019 and 2022. Daily, weekly, and monthly averages were plotted, and smoothed trends were illustrated using Locally Estimated Scatterplot Smoothing (LOESS) regression. Additionally, an animation of AOD data was created to depict temporal variations over the study period.

Linear regression models were developed to investigate the relationship between PM_2.5 and PM₁₀ concentrations. The models were evaluated for normality of residuals, homoscedasticity, and goodness of fit using appropriate statistical tests such as Shapiro-Wilk, Anderson-Darling, and Breusch-Pagan tests. To ensure the robustness of the regression results, three outliers were excluded from the analysis. These outliers were identified using Cook’s distance, a measure used to detect influential data points. Specifically, observations with Cook’s distance values exceeding a threshold, calculated as ${\displaystyle \frac{2}{n-k-1}}$ where n is the number of observations and k is the number of predictors in the model, were considered as outliers and subsequently removed from the regression analysis.

Geospatial analyses were conducted to map the distribution of PM10 and PM2.5 concentrations across the study area. Geographic coordinates of the sampling sites were used to create spatial plots. Spatial autocorrelation analysis was performed using Moran’s I to assess the degree of clustering or dispersion of particulate matter concentrations. Multiple approaches for defining spatial weights were tested, including distance-based and k-nearest neighbor methods, selecting the 5-nearest neighbors as the most appropriate spatial weighting scheme. This approach facilitated the identification of significant hotspots, influential sites and spatial outliers. Local Indicators of Spatial Association (LISA) were applied to visualize clusters of high and low values, while multiple custom functions were developed to detect influential sites, evaluate the strength of spatial relationships, and quantify spatial outliers based on Cook’s distance and Moran scatterplots [60,61,62,63].

To identify the meteorological and environmental variables that predict particulate matter (PM₁₀ and PM_2.5) concentrations, linear regression models were fitted. The predictor variables considered in this analysis included air temperature, air pressure, wind speed, rainfall, and Aerosol Optical Depth (AOD). All variables were obtained or derived from meteorological records and remote sensing products. Meteorological data, including air temperature, air pressure, wind speed, and rainfall, were sourced from the RDSD meteorological station, where daily averages were computed from high-frequency measurements to ensure consistency with particulate matter sampling periods. Aerosol Optical Depth (AOD) data were retrieved from the ’MCD19A2.061: Terra & Aqua MAIAC Land Aerosol Optical Depth Daily 1km’ product via Google Earth Engine [64], where ’MAIAC’ stands for Multi-angle Implementation of Atmospheric Correction. The data were extracted for an Area of Interest (AOI) encompassing the National District and an adjacent buffer zone. The resulting raster datasets were processed to obtain spatially averaged values at different temporal scales (daily, weekly, and monthly), ensuring comparability across sites. Before fitting the regression models, as mentioned before, spatial independence of the observations was verified using Moran’s I statistic.

The linear models were fitted separately for each period and particulate matter type. For 2019, models were evaluated during the July-August, September-October, and November-December periods. For 2022, the analysis considered the entire year for both PM_2.5 and PM₁₀ concentrations. Each model included the predictor variables as independent terms and the particulate matter concentrations as the dependent variable. The explanatory power of each model was assessed using the coefficient of determination ($R^2$), while the statistical significance of each predictor variable was evaluated through p-values. A tile plot was used to visually represent the relationships between predictor variables and particulate matter concentrations

All analyses were conducted using R statistical software [65]. Packages such as tidyverse, readxl, sf, sf, zoo, spatialreg, spdep, raster, terra, stars, forecast, caret, corrplot, GGally and gganimate, were utilized for data manipulation, statistical analysis, and visualization [55,66,67,68,69,70,71,72,73,74,75,76,77,78,79]. Custom functions were developed for specific tasks, such as performing correlation tests and fitting linear models.

3. Results

The basic statistics for particulate matter (PM) concentrations are shown in Table 1 and Figure 2. The mean PM₁₀ concentration in 2019 was 38.14 $\mu g/m^3$ (N = 26), while in 2022, the mean concentrations were 30.37 $\mu g/m^3$ for PM_2.5 (N = 30) and 62.18 $\mu g/m^3$ for PM₁₀ (N = 30). The Shapiro-Wilk test indicated that the data for 2022 significantly deviated from normality (p < 0.01), whereas the 2019 PM₁₀ data showed no significant deviation (p = 0.08).

The paired t-test comparing PM₁₀ concentrations between 2019 and 2022 indicated a significant difference, with a mean difference of -18.38 $\mu g/m^3$ (t = -2.38, p = 0.029), indicating that PM₁₀ concentrations were higher in 2022 than in 2019, with a 95% confidence interval ranging from -34.66 to -2.10 $\mu g/m^3$. In contrast, the Wilcoxon signed-rank test showed no statistically significant difference in the medians (V = 41, p = 0.054), although the result was close to the threshold for significance.

Turning to the data for 2022, the normality tests indicated that both PM_2.5 and PM₁₀ concentrations deviated significantly from normality (Table 2). Due to this deviation, Spearman’s rank correlation coefficient ($\rho$) was used to assess the relationship between the two pollutants. A moderate, but significant, positive correlation was observed between PM_2.5 and PM₁₀ concentrations, suggesting that higher levels of one pollutant were generally associated with higher levels of the other during this period.

To further explore the relationship between PM_2.5 and PM₁₀ concentrations in 2022, a linear regression model was fitted, excluding three outliers identified by Cook’s distance. The model indicated a significant positive relationship between the two pollutants, with PM_2.5 serving as a strong predictor of PM₁₀ levels. The regression equation was $y=29+0.92·x$, where y represents PM₁₀ and x represents PM_2.5. The model explains approximately 65% of the variance in PM₁₀ concentrations (R² = 0.65, adjusted R² = 0.63, p < 0.001). The residual standard error was 13.99, and the fitted model exhibited no significant violations of the assumptions of normality or homoscedasticity. This suggests that, after excluding the identified outliers, there is a robust linear association between PM_2.5 and PM₁₀ in 2022 (Figure 3).

Figure 5 shows the isopleths of the distribution around the region of PM₁₀ and PM_2.5 measured in Santo Domingo during both campaigns, 2019 and 2022 and the air quality stations (red circles). For the 2019 campaign, PM₁₀ concentrations were generally lower, with values ranging mainly between 10 and 25 ppm in the study area. Higher concentrations, exceeding 70 ppm, were observed in the westernmost part of the domain, as well as in some isolated spots in the north and center. Meanwhile, for the 2022 campaign, it is observed that for PM₁₀ it reaches values higher than 50 ppm distributed throughout almost the entire study domain with the exception of some isolated areas in the center, south, and southwest with lower values. However, for PM_2.5, the area with the highest values is located to the south of the study area, reaching values higher than 25 ppm.

The temporal trends of the weekly mean aerosol optical depth (AOD) for 2019 and 2022 are shown in Figure 4. Both years exhibit a clear seasonal pattern, with AOD values increasing steadily from January to a peak around July-August and subsequently declining toward December. In 2022, AOD values were consistently higher compared to 2019 throughout most of the year, particularly during the first half. This difference is more pronounced during the March to August period, where 2022 shows sharper increases, reaching maximum AOD values of approximately 0.4. The smoothed trends, visualized with LOESS regression, further emphasize these seasonal patterns, with confidence intervals highlighting significant deviations between the two years. Notably, AOD variability, as indicated by the error bars, is greater in 2019 during the middle of the year but becomes more stable in 2022. These results suggest a distinct difference in aerosol optical depth behavior between the two years.

The cross-correlation function (CCF) between weekly averaged AOD and PM₁₀ concentrations for 2019 and 2022 is presented in Figure 6. The analysis reveals a significant positive correlation at lag 0, suggesting that changes in AOD are contemporaneously associated with changes in PM₁₀. The correlation is strong at lag 0, with a coefficient close to 0.5, indicating a moderate association. At lag -1, the correlation is negative and relatively strong, while at lag -2, it becomes positive and strong. In contrast, the correlations at other lags, both positive and negative, are weak and not statistically significant. These findings indicate that variations in AOD and PM₁₀ concentrations occur synchronously on a weekly timescale, with minimal temporal displacement. This contemporaneous relationship highlights the potential of AOD as a reliable predictor for PM₁₀ concentrations when analyzed at a weekly resolution.

The local spatial autocorrelation analysis identified several influential sites and spatial outliers, along with a minimal number of hotspots, across different years and particulate matter (PM) sizes (Table 3 and Figure 7). However, as shown further, these localized patterns were not reflected in the global spatial autocorrelation analysis. For PM₁₀ in 2019, sites 8 and 22 were negatively influential, while site 23 was both positively influential and a positive spatial outlier, indicating higher PM₁₀ concentrations than expected. In 2022, site 10 for PM_2.5 was negatively influential and a negative spatial outlier, with sites 18 and 33 being positively influential; site 33 was also a positive spatial outlier and LISA hotspot, indicating a cluster of high PM_2.5 values. For PM₁₀ in 2022, site 29 was a LISA hotspot, and site 31 was negatively influential, revealing localized patterns of influence and clustering.

The spatial autocorrelation analysis for particulate matter (PM) using the Global Moran’s I statistic revealed non-significant results across all periods and particulate matter types assessed. Specifically, after log-transforming the particulate matter variables to better meet the assumption of normality, none of the Moran’s I tests yielded significant results. This indicates a lack of spatial dependence in the observed PM concentrations.

These results indicate that the particulate matter measurements do not exhibit significant spatial clustering or patterns, which aligns with the minimal number of LISA hotspots identified in previous analyses. This lack of spatial autocorrelation justifies the use of traditional statistical models that assume independent observations. Consequently, subsequent analyses can explore the relationships between PM levels and potential meteorological predictors without needing to account for any underlying spatial structure, thereby avoiding violations of the independence assumption.

The analysis of meteorological variables and Aerosol Optical Depth (AOD) as predictors of particulate matter concentrations revealed some interesting patterns. The results of the analysis were summarized visually in a tile plot, which highlights the significant predictor variables and their corresponding $R^2$ values (Figure 8). This figure enables a visual interpretation of the predictive strength of each variable across the analyzed periods and particulate matter types.

Notably, no significant predictors were identified for PM₁₀ levels during the months of July-August and November-December 2019 using linear models. However, the results indicated that for PM₁₀ concentrations during September-October 2019, wind speed emerged as a significant predictor, albeit with a relatively low explanatory power (R²=0.181).

In contrast, for the year 2022, multiple variables showed predictive power for both PM_2.5 and PM₁₀ levels. Specifically, air pressure was identified as a significant predictor for PM_2.5, although with moderate explanatory power (R²=0.361). For PM₁₀ in 2022, both AOD and air pressure exhibited strong predictive capabilities, with the model achieving a high R² of 0.752. Additionally, rainfall was also a significant predictor for PM₁₀ during this period, further indicating that meteorological conditions played a more influential role in explaining variations in particulate matter concentrations in 2022 compared to 2019.

4. Discussion

This study evaluated PM₁₀ and PM_2.5 concentrations in Santo Domingo, focusing on their spatial distribution, temporal variability, and relationships with meteorological and environmental variables. The findings reveal significant insights into particulate matter dynamics, particularly the role of aerosol optical depth (AOD), meteorological predictors, and localized spatial patterns.

The observed differences in PM concentrations between 2019 and 2022 can be attributed to seasonal variability and environmental factors. Notably, higher PM₁₀ levels in 2022 coincided with the mid-year months (May–July), characterized by increased Saharan dust activity [80,81]. This aligns with the elevated AOD values during this period, as highlighted by the LOESS regression trends. In contrast, the lower PM₁₀ concentrations in 2019, sampled between July and December, may reflect reduced dust transport and increased precipitation, which likely enhanced particle deposition.

Despite a substantial increase in the national vehicle fleet from 2019 to 2022 [82], the minimal growth in the National District and Santo Domingo province suggests that vehicular emissions are unlikely to explain the observed PM₁₀ increase in 2022. Instead, meteorological conditions such as air pressure and rainfall, alongside AOD values (likely correlated with higher Saharan air layer concentrations), emerged as significant predictors, as demonstrated by the regression models. The strong explanatory power of these variables, particularly for PM₁₀ (R² = 0.752), underscores their critical role in influencing particulate matter levels.

The cross-correlation analysis revealed a synchronous relationship between AOD and PM₁₀ at lag 0, indicating that real-time AOD data could serve as a reliable proxy for estimating PM₁₀ concentrations, consistent with findings from previous studies that demonstrated strong correlations between AOD and surface particulate matter levels [59,83,84]. The temporal alignment suggests that satellite-derived AOD products could enhance air quality monitoring in regions with limited ground-based observations, as supported by studies showing that integrating AOD into air quality models significantly improves PM₁₀ and PM_2.5 estimations, particularly in areas with sparse monitoring networks [85,86,87,88,89,90,91,92].

Spatial autocorrelation analysis identified localized patterns, including influential sites, hotspots, and spatial outliers. For instance, site 23 in 2019 exhibited high PM₁₀ concentrations, serving as both a positive spatial outlier and an influential site. In 2022, site 33 was a LISA hotspot for PM_2.5, with significant clustering of high values. However, the absence of global spatial autocorrelation suggests that PM concentrations are not driven by broader spatial patterns, allowing the use of traditional regression models that assume independent observations. This lack of spatial autocorrelation is expected when observations are taken on different dates, particularly for particulate matter concentrations, which represent a highly dynamic phenomenon strongly influenced by atmospheric conditions [93,94]. This further underscores the need to model its behavior using physical atmospheric models to complement statistical approaches [95,96].

These findings emphasize the importance of integrating spatial and temporal analyses to uncover localized phenomena and validate predictive models. While the study highlights key predictors and spatial patterns, future research should explore the influence of land use, industrial activities, and local emission sources to develop a more comprehensive understanding of air quality dynamics in Santo Domingo.

Based on the findings of the study, which highlight the role of meteorological factors, urban activities, and aerosol concentrations, the following policy recommendations and mitigation strategies can help improve air quality in Santo Domingo: strengthening air quality monitoring and data integration, reducing vehicular emissions, controlling industrial and construction emissions, implementing urban planning for better air quality, promoting public awareness and community engagement, and enhancing policy coordination and enforcement.

5. Conclusions

This study provides a comparative assessment of PM₁₀ and PM_2.5 concentrations in Santo Domingo for 2019 and 2022, offering key insights into their temporal trends, meteorological predictors, and spatial variability. The significant differences in particulate matter levels between the two years are closely linked to seasonal meteorological conditions and Saharan dust activity, rather than local anthropogenic sources.

The regression models identified AOD, air pressure, and rainfall as robust predictors of PM concentrations, highlighting the potential of integrating remote sensing data with ground-based measurements to enhance air quality monitoring. The cross-correlation analysis further supports the use of AOD as a real-time proxy for PM₁₀, providing valuable tools for policymakers and researchers.

Spatial analyses revealed localized influences, with certain sites demonstrating significant clustering and outlier behavior. However, the lack of global spatial autocorrelation underscores the need for targeted interventions rather than broad regional strategies. The study underscores the importance of incorporating localized meteorological and environmental data into air quality management plans.

Future studies should aim to expand the temporal and spatial scope of sampling, incorporate high-resolution emission inventories, and assess the health implications of observed PM concentrations. These efforts will enhance the ability to design effective mitigation strategies and improve urban air quality in Santo Domingo and similar urban environments.

6. Recommendations

While this study provides a comparative assessment of PM₁₀ and PM_2.5 levels in different areas of Santo Domingo, future research should incorporate continuous measurements or repeated sampling to assess temporal variability with greater precision. To improve continuity in future studies, it is recommended to establish a fixed monitoring network to analyze the evolution of pollutants at the same sites over time.

The results of this study suggest that, in addition to meteorological factors, anthropogenic sources such as traffic and industrial activity may play a significant role in PM levels in Santo Domingo. Future research should include detailed emission inventories and impact analyses of traffic control policies and industrial regulations to better understand their influence. Implementing mitigation strategies will require a multi-sectoral approach involving government agencies, industries, academia, and local communities. By integrating better monitoring, stricter regulations, sustainable transportation, and urban greening, Santo Domingo can significantly reduce air pollution and improve public health.