Heavy Metal Concentrations in Particulate Matter: A Case Study from Santo Domingo, Dominican Republic, 2022

1. Introduction

Air pollution is a critical environmental and public health issue, particularly in urban areas where anthropogenic activities contribute to elevated levels of harmful pollutants [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16]. Among these pollutants, particulate matter (PM) is of special concern due to its ability to carry various toxic substances, including heavy metals. PM comprises tiny solid particles and liquid aerosols containing acids, organic compounds, metals, and dust [17,18,19,20,21,22,23,24]. These particles originate from natural sources, such as wildfires, volcanic activity, and Saharan dust transport, as well as human activities like industrial emissions, vehicle exhaust, and fossil fuel combustion.

Heavy metals are generally defined as elements with high atomic weight and a density at least five times greater than that of water. Metalloids, such as arsenic (As), exhibit characteristics intermediate between metals and non-metals and are often grouped with heavy metals due to similar toxicological behaviour [25]. Some heavy metals, including iron (Fe), copper (Cu), zinc (Zn), manganese (Mn), cobalt (Co), and nickel (Ni), are considered essential trace elements for biological functions, participating in enzymatic processes, redox reactions, and cellular signalling [26]. However, even essential metals can become toxic when present in elevated concentrations, while non-essential elements like cadmium (Cd), lead (Pb), mercury (Hg), vanadium (V), and arsenic (As) are toxic even at low levels [27].

Heavy metals such as As, Mn, Cu, Cd, Pb, Zn, Ni, V, and Co are frequently detected in particulate matter (PM), especially in its respirable fractions, PM_2.5 and PM₁₀. These elements can bind to fine and coarse particles and are released into the atmosphere through both anthropogenic sources—including metallurgy, combustion processes, vehicular traffic, waste incineration, and industrial activity—and natural events such as volcanic eruptions and wind-driven dust resuspension [28].

In the PM₁₀ fraction, heavy metals are predominantly associated with coarser particles that tend to deposit in the upper respiratory tract. In contrast, PM_2.5 represents a greater health hazard, as fine particles can penetrate deep into the alveolar regions of the lungs and enter the bloodstream, carrying toxic metals with them [29]. These metals are known not only for their capacity to bioaccumulate and contribute to respiratory and cardiovascular diseases, but also for their potential to induce genotoxic and cytotoxic effects [13,30]. Due to their greater surface area and reactivity, fine particles are especially efficient at penetrating biological barriers and triggering oxidative stress, DNA damage, inflammation, and apoptosis in human cells, particularly in pulmonary and epithelial tissues [25,31,32,33,34,35,36]. Beyond human health, heavy metals in PM can also impact terrestrial and aquatic ecosystems. Once deposited, these metals may accumulate in soils and sediments, become absorbed by vegetation, and enter the food chain, where they may biomagnify and affect biodiversity and ecological functions [37].

Urban areas in developing countries are especially vulnerable to air pollution, as rapid urbanization and industrialization often outpace environmental regulations [3,7,8,11,14,15,38,39]. The city of Santo Domingo, located in the National District on the south-central coast of the Dominican Republic, exemplifies these challenges. While the city’s diverse urban environments provide an opportunity to study the dynamics of air pollution, limited research has been conducted to investigate the concentrations and sources of heavy metals in PM [40,41,42,43].

The atmospheric deposition of heavy metals is influenced by various factors, including wind patterns, precipitation, and local emission sources. Understanding the concentrations and atmospheric fluxes of these metals is essential to assess their environmental and health impacts. Despite its importance, data on heavy metal concentrations in PM from Santo Domingo remain scarce, underscoring the need for targeted studies [40,41].

This study focuses on the analysis of heavy metal concentrations in PM_2.5 and PM₁₀ collected during a sampling campaign in Santo Domingo in 2022. The primary objectives are: (1) to evaluate the concentrations of As, Cd, Cr, Cu, Fe, Hg, Mn, Ni, Pb, V and Zn in PM fractions; (2) to assess imputation methods for left-censored values to extract meaningful conclusions about metal concentrations and their variability; and (3) to provide insights into the potential sources and environmental implications of heavy metal deposition in the region, as well as their distribution across sampling sites.

By providing a comprehensive analysis of heavy metal concentrations in PM, this research contributes to a deeper understanding of air pollution dynamics in Santo Domingo and offers a scientific basis for policy interventions aimed at mitigating environmental and public health risks.

2. Materials and Methods

The study was carried out in Santo Domingo, National District, Dominican Republic (approx. $18.{49}^{\circ }\mathrm{N},69.{96}^{\circ }\mathrm{W}$), covering a range of urban environments with different land uses (Figure 1). The sampling locations included public and private schools, a university campus, and a major urban park, following the initial site selection proposed by Caballero-González. A total of 30 air quality samples were collected between May and July 2022 for both PM_2.5 and PM₁₀, with one sample per site due to logistical and budgetary constraints. Although the measurements of PM mass concentrations (expressed in $\mu$g/m³) are not presented in this study, they are available in the work of Matos-Espinosa et al..

This single-day-per-site design, adopted for feasibility reasons, inherently restricts the ability to assess intra-site temporal variability or capture seasonal trends, but it offers a practical approach for characterizing the spatial distribution of particulate matter over a broad urban area. Single-day measurements, when distributed systematically across representative sites within a narrow temporal window (May to July), can still provide meaningful spatial comparisons, especially when meteorological conditions remain relatively stable throughout the sampling period [44]. Furthermore, by ensuring that sampling activities occurred on consecutive or near-consecutive days during each week, the resulting dataset supports the computation of time-averaged indicators and minimizes potential biases introduced by sporadic or asynchronous sampling. Although this approach does not replace long-term monitoring, it aligns with widely used methods in resource-constrained settings and has been shown to yield valuable insights into spatial pollution patterns across urban environments [44].

Particulate matter was collected using MiniVol^TM TAS Portable Air Samplers (AirMetrics Co., Eugene, OR, USA) [45,46]. These devices operate by drawing air through size-selective impactors to separate PM₁₀ and PM_2.5 fractions based on aerodynamic diameter, following the principle of inertial impaction. Gravimetric analysis was conducted using pre-weighed 47 mm PTFE (polytetrafluoroethylene) filters, chosen for their low hygroscopicity, chemical inertness, and stable mass under conditioning, which make them suitable for mass determination and subsequent chemical analysis. Each sampler was operated for 24 h at a nominal flow rate of 5.0 L/min ($\pm 5\%$), with an estimated detection limit of 3–5 µg/m³ and a reported measurement accuracy within $\pm 10\%$, assuming proper maintenance and calibration. The MiniVol^TM samplers are limited to PM₁₀ and PM_2.5 and do not support high-frequency measurements or PM₁ collection, restricting the analysis to these two common size fractions.

The devices were installed at a height of 1.5 to 3 m above ground level, either on flat rooftops or securely positioned at ground level, ensuring consistency across sites to allow for comparability. The sampling strategy followed a systematic layout, ensuring an approximate 2 km separation between sites. The final selection of specific sampling points, including educational and recreational areas, was refined using the nearest-neighbor approach, optimizing their placement based on predefined ideal locations determined through spatial analysis software.

Heavy metal concentrations in PM_2.5 and PM₁₀ were determined by energy-dispersive X-ray fluorescence spectrometry (EDXRF). Analyses were performed using a Skyray Instruments EDX3600B spectrometer equipped with a Si(Li) detector positioned at a ${45}^{\circ }$ angle relative to the Ag anode X-ray source. The excitation parameters were set at 40 kV and 600 $\mu$A. Total concentrations of Fe, Mn, Cr, Cu, Ni, Zn, Pb, and As were quantified based on calibration curves obtained from certified Standard Reference Materials (SRMs). The quality of each calibration was assessed through the coefficient of determination ($R^2$), which ranged from 0.990 to 0.999. Elemental intensities were processed using the proprietary Skyray Instruments software, provided by the manufacturer [47,48].

Although energy-dispersive X-ray fluorescence (EDXRF) has relatively high detection limits compared to more sensitive techniques such as ICP-MS, it was selected for this study due to several practical and scientific advantages. EDXRF is a non-destructive, cost-effective, and rapid analytical method that allows for the simultaneous quantification of multiple elements directly on filter media without the need for complex sample preparation. These features are particularly valuable in resource-constrained settings or large-scale environmental monitoring campaigns, where budget and time limitations must be balanced with data quality. Moreover, EDXRF provides sufficient accuracy and reproducibility for detecting moderate to high concentrations of metals commonly found in urban particulate matter. The dataset includes measurements of arsenic (As), cadmium (Cd), chromium (Cr), copper (Cu), iron (Fe), mercury (Hg), manganese (Mn), nickel (Ni), lead (Pb), vanadium (V), and zinc (Zn), with reported values in parts per million (ppm). Samples with metal concentrations below the instrument’s detection limit were treated as left-censored data. This limitation, which affected elements such as As, Mn, and Hg, was addressed through the application of robust statistical methods for censored data, enabling meaningful interpretation without compromising the integrity of the dataset.

Figure 1. Sampling sites for particulate matter (PM_2.5 and PM₁₀) in the National District of Santo Domingo during 2022. Site names and classifications are shown in the list to the right. Geographic coordinates for each site are provided in Table A1 and Table A2.

Figure 1. Sampling sites for particulate matter (PM_2.5 and PM₁₀) in the National District of Santo Domingo during 2022. Site names and classifications are shown in the list to the right. Geographic coordinates for each site are provided in Table A1 and Table A2.

All analyses were conducted using R statistical environment (Version 4.4.0) [49], using a combination of packages, including tidyverse for data wrangling, gstat and automap for spatial interpolation of metal concentrations, NADA for censored data imputation, pcaMethods for PCA, and ggplot2 for visualization [50,51,52,53,54,55,56]. The complete workflow was implemented in an R Markdown document for reproducibility.

Initial data exploration included calculating summary statistics and correlation analyses among heavy metal concentrations in PM_2.5 and PM₁₀ (Table A1 and Table A2). However, due to a high proportion of left-censored values in several metals, correlation analyses were not used for further interpretation.

Summary statistics were computed before and after imputation to assess the central tendency and variability of metal concentrations. Metals with more than 50% left-censored values were excluded from the imputation process, following best practices in environmental data analysis [57].

To visualize the spatial distribution of heavy metals in PM_2.5 and PM₁₀, we applied geostatistical interpolation using ordinary kriging [58,59,60,61]. Only those metals with at least 50% valid observations were included in the analysis. The monitoring sites were converted into an sf object in R and reprojected to UTM Zone 19N (EPSG:32619) [62,63]. A regular interpolation grid was created with a resolution of 50 meters, extending the bounding box of the monitoring points by 2 kilometers to reduce edge effects.

The automap R package was used to automatically fit variogram models (spherical or exponential) based on the smallest residual sum of squares with respect to the sample variogram, and perform kriging for each metal independently [53,58,59,60,61]. Interpolated values were transformed into a raster and then used to generate filled contour levels and smoothed isolines. The individual maps for each metal were assembled into a single panel using the ggplot2 and patchwork R packages [64,65]. The shaded polygon in the background of each map represents the official administrative boundary of the National District of Santo Domingo.

The instrumental detection limits (IDLs) for heavy metals in PM_2.5 and PM₁₀ were determined following established protocols for X-ray fluorescence spectrometry [66,67] as well as using empirical laboratory measurements. The IDLs were derived from multiple measurements of certified reference materials (CRMs), specifically SRM-NIST-IAEA samples, analyzed under the same conditions as the field samples. The detection limit for each metal was established based on the lowest concentration that could be reliably distinguished from background noise, considering the variability across multiple replicates [68]. The IDLs values for elements such as Al, Si, P, S, Mn, Fe, Ni, Cu, Zn, As, Br, Sr, Cd, and Pb were compiled in an accompanying reference table, ensuring consistency in the reporting of metal concentrations [69,70]. Values below these detection limits were treated as left-censored and were subsequently handled using statistical imputation methods where appropriate [57].

For metals with less than 50% of measurements below detection limits, imputation was performed using regression on order statistics (ROS) implemented in the NADA package in R statistical environment [55]. The ROS method estimates censored values based on the empirical distribution of detected values, providing a statistically sound approach for handling non-detects while avoiding strong parametric assumptions [57].

Importantly, imputed values were only used for global descriptive statistics and not for multivariate analyses, such as principal component analysis (PCA), where preserving the original data structure was required.

PCA was applied separately to PM_2.5 and PM₁₀ metal concentrations to explore their distribution across sampling sites and identify potential sources. Only metals with sufficient observations (i.e., limited missing values) were included in the PCA. The final selection of metals for PCA was based on the presence of complete cases across sites, ensuring the robustness of the analysis. The PCA was performed using the pca function from the pcaMethods package in R, employing the non-linear iterative partial least squares (NIPALS) algorithm [71,72]. This method is particularly suitable for datasets with missing values, as it iteratively estimates principal components while handling incomplete data without requiring listwise deletion. The input dataset was standardized (mean-centered and scaled to unit variance) before PCA computation.

The results were visualized using biplots to examine the relationships between metals and sampling sites. The biplots were generated using ggplot2, with individual observations plotted in the principal component space and metal loadings represented as vectors scaled for interpretability. The explained variance of each principal component was assessed to determine the number of meaningful components.

Although detailed external datasets such as traffic counts or land-use inventories were not available for integration in this study, the interpretation of principal components was informed by established metal source profiles and contextual knowledge of the urban environment. Metals such as Cu, Zn, and Pb, which loaded heavily on the first component, are commonly associated with vehicular emissions and mechanical wear, while Fe and V suggest contributions from industrial activities or resuspended soil. Future research will aim to incorporate high-resolution auxiliary data, including traffic density, industrial zoning, and meteorological parameters, to further refine source attribution and strengthen causal inferences.

While the toxicological relevance of heavy metals in particulate matter is well established, this study did not include a formal health risk assessment or estimates of population exposure due to the unavailability of local epidemiological and health outcome data. Our primary objective was to establish a baseline of metal concentrations across urban sites, which is a necessary first step toward future risk-based evaluations. The spatial patterns and concentration levels reported here can serve as input for subsequent studies focused on health impact modelling, particularly those incorporating population vulnerability, exposure duration, and toxicological benchmarks. We recognize the importance of such analyses and recommend them as a critical next step in advancing air quality management in the region.

3. Results

After applying regression on order statistics (ROS) to impute left-censored values, descriptive statistics for metal concentrations in PM_2.5 were obtained. Manganese (Mn) and Mercury (Hg) were excluded from the imputation process since more than 50% of their values were censored, making reliable estimation unfeasible. The results indicate substantial variability among metals, with mean concentrations ranging from 2.48 ppm for As to 21.89 ppm for Cu (Table 1). Among the analyzed elements, Cu and Zn exhibited the highest median values (22.00 ppm and 19.80 ppm, respectively), whereas As and Cd had the lowest medians (2.40 ppm and 2.50 ppm, respectively). The standard deviation values reveal notable dispersion, particularly for V (9.18 ppm) and Fe (6.10 ppm), indicating significant variability across the sampling sites. Additionally, the standard error values, which provide an estimate of uncertainty in the mean concentrations, were lowest for Zn (0.16 ppm) and highest for V (1.68 ppm), suggesting more stable concentration estimates for Zn compared to highly variable metals like V.

The proportion of censored values varied across metals, with the highest number observed for V (15 censored values) and Fe (14 censored values), while Cu and Zn had no censored values. This suggests that certain metals frequently fell below the instrumental detection limits, requiring imputation to obtain meaningful statistical summaries. The highest recorded metal concentration was 36.9 ppm for V, while the lowest detected value was 0.58 ppm for Cd. Figure 4 illustrates the mean concentrations along with their standard errors, highlighting the metals with greater measurement uncertainty. The error bars emphasize the variation in concentration levels across sites, with Fe and V showing the most considerable spread, reinforcing their observed variability in the dataset.

The analysis of metal concentrations in PM₁₀ revealed differences in both the magnitude and variability of detected levels. Copper (Cu) and Zinc (Zn) exhibited the highest mean concentrations, with values of 21.15 ppm and 20.14 ppm, respectively, while Cadmium (Cd) and Chromium (Cr) showed the lowest means at 3.79 ppm and 3.89 ppm. Vanadium (V), despite having a relatively moderate mean concentration (8.61 ppm), displayed the highest observed value among all metals (43.6 ppm), suggesting a highly skewed distribution. Similarly, Iron (Fe) exhibited substantial variability, with concentrations ranging from 3.41 ppm to 28.4 ppm. It is important to note that Manganese (Mn), Mercury (Hg), and Arsenic (As) were excluded from imputation due to excessive censoring, as more than 50% of their values were below detection limits, making statistical estimation unreliable. The variability in concentrations is further illustrated in Table 2.

Figure 2. Spatial interpolation of metal concentrations (ppm) in PM_2.5 particles across the study area using ordinary kriging. Each panel corresponds to a different metal: As, Cd, Cr, Cu, Fe, Ni, Pb, V, and Zn. Filled contour maps show estimated concentration gradients, with isolines indicating interpolated concentration levels. The shaded polygon in the background represents the boundary of the Distrito Nacional of Santo Domingo. Scale bars and north arrows are provided for spatial reference.

Figure 2. Spatial interpolation of metal concentrations (ppm) in PM_2.5 particles across the study area using ordinary kriging. Each panel corresponds to a different metal: As, Cd, Cr, Cu, Fe, Ni, Pb, V, and Zn. Filled contour maps show estimated concentration gradients, with isolines indicating interpolated concentration levels. The shaded polygon in the background represents the boundary of the Distrito Nacional of Santo Domingo. Scale bars and north arrows are provided for spatial reference.

When considering the uncertainty associated with these estimations, the standard error values indicate that Zn had the most stable concentration estimates (0.13 ppm), whereas V had the highest level of uncertainty (1.88 ppm), reflecting its large dispersion. The extent of censored values also varied significantly among metals, with Fe (19 censored values) and V (16 censored values) being the most affected. In contrast, Cu, Zn, and Ni had no censored values, implying that their concentrations consistently exceeded detection limits. The graphical representation in Figure 5 highlights the differences in concentration variability across metals, emphasizing the higher dispersion in V and Fe, which suggests spatial heterogeneity in their deposition patterns.

The principal component analysis (PCA) for metal concentrations in PM_2.5 (Figure 6) reveals the distribution of sampling sites along the first two principal components, which together explain a substantial proportion of the variance in the dataset. The first component (PC1) accounts for 28.3% of the variance, while the second component (PC2) explains 18.6%. Sampling sites such as the Agrarian Institute of Dominican Republic, APEC University, and San Judas Tadeo School are positioned at the extremes along PC1, whereas sites like the American School of Santo Domingo, Victor Estrella Liz School, and Prof. Adolfo González School are located towards the center. The loading plot shows that metals such as Zn, V, and Pb contribute strongly to the first component, while Fe and Cr have a more substantial influence along the second component.

Figure 3. Spatial interpolation of metal concentrations (ppm) in PM₁₀ particles across the study area using ordinary kriging. Each panel corresponds to a different metal: Cd, Cr, Cu, Fe, Ni, Pb, V, and Zn. Filled contour maps show estimated concentration gradients, with isolines indicating interpolated concentration levels. The shaded polygon in the background represents the boundary of the Distrito Nacional of Santo Domingo. Scale bars and north arrows are provided for spatial reference.

Figure 3. Spatial interpolation of metal concentrations (ppm) in PM₁₀ particles across the study area using ordinary kriging. Each panel corresponds to a different metal: Cd, Cr, Cu, Fe, Ni, Pb, V, and Zn. Filled contour maps show estimated concentration gradients, with isolines indicating interpolated concentration levels. The shaded polygon in the background represents the boundary of the Distrito Nacional of Santo Domingo. Scale bars and north arrows are provided for spatial reference.

Figure 4. Mean Concentration of Metals in PM_2.5 with Standard Error

Figure 4. Mean Concentration of Metals in PM_2.5 with Standard Error

Figure 5. Mean Concentration of Metals in PM₁₀ with Standard Error

Figure 5. Mean Concentration of Metals in PM₁₀ with Standard Error

Figure 6. PCA Biplot, metals in PM_2.5

Figure 6. PCA Biplot, metals in PM_2.5

In the case of PM₁₀ (Figure 7), the first two principal components explain 28.5% and 22.4% of the variance, respectively. The PCA plot reveals a clustering pattern where sites such as Aida Cartagena Portalatín School and The Community for Learning are positioned towards the positive extremes of PC1, while Los Prados School, María Auxiliadora School, and Notre Dame School are situated towards the negative end. The loading vectors indicate that Zn, V, and Fe show a strong correlation with PC1, whereas Pb, Cr, and Ni have more influence along PC2. Compared to PM_2.5, the PCA for PM₁₀ shows a slightly different spatial distribution of sites, suggesting variations in the metal concentration patterns.

Across both analyses, certain sites exhibit distinct positions within the PCA space, indicating differences in their metal concentration profiles. For instance, Capotillo School and Arroyo Hondo School appear centrally located in both analyses, while Salomé Ureña School and República Dominicana School occupy different relative positions between PM_2.5 and PM₁₀. The biplots further illustrate the contribution of individual metals, where V, Cr, and Pb consistently appear as key contributors to the variability in both fractions. The separation among sites along PC1 and PC2 suggests underlying differences in metal distributions across urban environments.

Figure 7. PCA Biplot, metals in PM₁₀

Figure 7. PCA Biplot, metals in PM₁₀

4. Discussion

This study aimed to (1) quantify the concentrations of selected heavy metals in PM_2.5 and PM₁₀ in Santo Domingo, (2) apply statistical imputation techniques to handle left-censored data, and (3) explore the spatial distribution and potential sources of these metals through principal component analysis. These objectives have been met successfully: the analysis provided detailed concentration profiles for nine metals, robustly addressed censoring through ROS, and identified spatial variability and possible emission sources using PCA. The study contributes new baseline data on airborne metal pollution in the capital city of the Dominican Republic, a context where environmental monitoring efforts remain limited [44]. The findings provide a foundation for both regulatory interventions and future scientific research.

The high concentrations of Cu and Zn across both PM fractions suggest dominant and persistent anthropogenic sources, possibly related to vehicular traffic, brake and tire wear, and mechanical workshops [73,74]. These metals are commonly found in urban PM worldwide and their consistent detection in all sites supports their ubiquity in the city’s atmosphere [75,76]. The presence of Pb and Ni, although at moderate levels, remains concerning due to their toxicity and potential for chronic exposure. Leaded gasoline was phased out in the region decades ago, but legacy pollution from past vehicular emissions and industrial activity may still influence current concentrations [77,78,79].

Vanadium (V) and iron (Fe) exhibited high variability, particularly in PM₁₀, suggesting heterogeneous sources or resuspension processes [80,81]. V is often associated with the combustion of heavy fuels, such as bunker oil used in marine transport and some electricity generation facilities [82,83]. The elevated standard errors and extreme values observed for V may reflect occasional long-range transport or local emission peaks. Similarly, the broad concentration range of Fe may indicate contributions from soil resuspension, construction activities, or industrial dust [80,84].

The PCA results revealed patterns of co-occurrence among metals, allowing for tentative source attribution. In both PM_2.5 and PM₁₀, PC1 was strongly influenced by Zn, Cu, and V, suggesting a common source, likely traffic-related or industrial combustion [83,84]. Site differentiation along the PCA axes implies that exposure to metal pollutants is spatially uneven, with educational centers like APEC University and Aida Cartagena Portalatín School showing distinct signatures. Such variation may correspond to micro-environmental differences, such as traffic density, building layout, and wind flow dynamics [85,86].

This study has several limitations. First, the number of samples (30) constrains the statistical power and generalizability of the findings. Although the sampling design was spatially optimized, temporal variability was not captured, as all samples were collected during a single campaign. Additionally, the exclusion of several metals (e.g., Mn, Hg, As) from the PCA due to high censoring limits the scope of source identification. The use of XRF, while cost-effective and non-destructive, also has higher detection limits compared to other analytical methods such as ICP-MS, potentially leading to underestimation of trace metals [69,87].

These results prompt several new research questions. Future studies should investigate seasonal and diurnal patterns in metal concentrations, assess deposition rates to soil and vegetation, and explore health outcomes associated with chronic exposure in children and sensitive populations. Integrating land use regression (LUR) models or chemical mass balance (CMB) models would help refine source attribution [88,89]. Moreover, monitoring metals not captured in this study, including those associated with electronic waste and emerging contaminants, could yield insights into evolving pollution sources in urban contexts.

5. Conclusions

This work provides a foundational assessment of heavy metals in atmospheric particulate matter in Santo Domingo. It reveals the presence of persistent pollutants, spatial heterogeneity in exposure, and key indicators of anthropogenic activity. The study underscores the need for expanded air quality monitoring and targeted mitigation strategies to protect public health and environmental quality in the region.