Contribución y evolución de las fuentes de material particulado: (en PM10) en el puerto de Alicante durante el periodo 2017-2023

• Autores: Daniel Tobarra Moroño
• Directores de la Tesis: Adoración Carratalá Gimenez (dir. tes.), Eduardo Yubero Funes (dir. tes.)
• Lectura: En la Universitat d'Alacant / Universidad de Alicante ( España ) en 2025
• Idioma: español
• Número de páginas: 157
• Tribunal Calificador de la Tesis: Franco Lucarelli (presid.), Nuria Ortuño García (secret.), Jaime Javier Crespo Mira (voc.)
• Programa de doctorado: Programa de Doctorado en Ingeniería Química por la Universidad de Alicante
• Enlaces
  • Tesis en acceso abierto en: RUA
• Resumen

  • Particulate Matter (PM) is usually classified based on its size, with particles having an aerodynamic diameter smaller than 10 μm (PM10) being the largest category frequently studied. PM10 are relevant as 10 μm is the maximum particle size considered to be able to enter the lungs. This size range also includes all the particles types that can penetrate the human body. Particles of natural origin, such as sea spray and particles coming during sand dust events, tend to be bigger and mostly collected as PM10, whereas anthropogenic particles tend to be smaller (< 2,5 μm). PM10 particles are a worldwide concern, with several studies highlighting their effects in human health, particularly in children, the elderly and individuals with respiratory diseases. A large number of studies in exposed human populations over decades have shown associations between prenatal exposure to fine particulate matter (PM) and adverse birth outcomes, respiratory disease, and impaired neurodevelopment. Most international organizations, such as the World Health Organization (WHO) and European Environment Agency (EEA), as well as national organizations like the United States Environmental Protection Agency (EPA), have established various and evolving limit values for key pollutants to protect public health. The WHO, for example, proposes different limits to be used as targets for reducing emissions to give countries a guide. Air quality monitoring is essential to detect exceedances of the PM10 daily limit set by the European Directive 2024/2881 (50 μg/m3 by 2030 and 20 thereafter), while the World Health Organization recommends a lower limit for PM10 of 15 μg/m3. In 2019, 99 % of the world population was living in areas where PM10 concentrations exceeded WHO recommendations. Accurate measurements are essential to assess whether the PM10 concentrations in each area comply with the established daily limits. Two different methods are used for this purpose: gravimetric samplers and automatic samplers. Gravimetric samplers collect physical samples and determine the concentration based on the mass difference of the filter before and after sampling, and the volume of the air sampled. Gravimetric samplers are the reference method due to their accuracy; but they require a significant amount of manual work to prepare and process the samples. On the other hand, automatic samplers determine PM10 concentrations by associating a physical property with the mass of the particulate matter. Among these, the samplers based on laser spectrometry, usually called Optical Particle Counters (OPCs), have become increasingly popular. The main advantage of these automatic samplers is their ability to operate autonomously and to provide real-time data, offering a much higher temporal resolution than gravimetric methods. Despite their advantages, the high cost of these devices makes large-scale deployment challenging, limiting their use in extensive monitoring networks. In an attempt to overcome this limitation, an evolution of OPC technology has been developed based on Low-Cost Sensors (LCS). These sensors operate on the same principle as OPCs but at a much lower costs, enabling the creation of High-Density Sensor Networks (HDSN). This cost reduction is achieved through the use of lower-quality components, simplified algorithms, and the absence of humidity control in the sampled air. Both, automated samplers and the Low-Cost Sensors require frequent calibration with gravimetric sensors to ensure their correct measurements. However, reducing PM10 concentration remains complex, as PM10 is an umbrella term covering particles of different origin and composition. Identifying the sources of these particles is essential to implement effective mitigation strategies. Chemical analysis of PM10 collected by gravimetric samplers, followed by statistical analysis using models like Positive Matrix Factorization (PMF), is one of the most widely used methods to identify and quantify PM10 sources. This quantitative methodology allows the application of evidence-based policymaking. In addition, this information can be used to study trends in individual sources affecting PM10 levels such as road traffic and maritime traffic. This type of studies provides the knowledge needed to assets the effectiveness of existing environmental policies and interventions over time. Sources affecting PM10 concentrations can be either local, long-range transport or hybrid. Local sources are highly dependent on the characteristics of the area and may include road traffic, local industries, port activities, airports, sea salt, agriculture or livestock, all of which can have a significant impact on concentrations. Long-range transport sources, on the other hand, originate from transboundary emissions, such as desert dust intrusions, shipping emissions, or certain anthropogenic pollutants transported over long distances. The aim of this doctoral thesis is to determine the PM10 concentrations in the Port of Alicante, a coastal area heavily influenced by a port. To achieve this, this study focuses on identifying the sources influencing the PM10, analyzing their evolution over time, and evaluating the impact of mitigations measures and regulations in the area. To achieve this objective, several specific sub-objectives have been defined. 1. Analyze the evolution of PM10 concentrations and exceedances of the PM10 daily limit value over the study period (2017 to 2023) to assess compliance with current legislation. This analysis will provide a general understanding of air quality trends in the area. 2. Investigate the local meteorological dynamics to identify frequent wind conditions. Study how wind regimes influence PM10 and NO2 concentrations and use high resolution hourly data to identify the main local sources of PM10. 3. Use chemical data from gravimetric samplers to identify the sources contributing to PM10 levels in the study area. This objective is further subdivided into: a. Determine which PM10 sources have the greatest impact on air quality in the region. b. Analyze the evolution of both natural and anthropogenic sources over a five-year period (2017-2021). i. First, by assessing the effects of a drastic and prolonged reduction in anthropogenic emissions during the COVID-19 lockdown on PM10 concentrations and their sources. ii. Second, by examining long-term concentration trends to evaluate the effectiveness of various mitigation measures and regulations implemented during this period. c. Harmonization of PM10 measurements from different OPC samplers located at the perimeter of the port of Alicante using a gravimetric sampler as a reference. To improve calibration, an intercalibration system for OPC samplers is proposed as an alternative to the current method, which relies solely on infrequent external calibration campaigns. This system aims to enable the use of calibrated hourly time series. The following sub-objectives guide this process: i. Determine the equivalence between the external gravimetric campaigns and the proposed reference stations, selecting the most appropriate station for long-term calibration. Generate a theoretical data series for each measurement point using the selected reference station. ii. Correct OPC data series using a multi-linear model that includes meteorological variables to reproduce the theoretical series. iii. Correct OPC data series using non-linear machine learning models, such as Random Forest, to reproduce the theoretical data series. The main sampling area of this study, Tinglado Frutero (TF), is located within the port of Alicante. In August 2017, three Optical Particle Counters (OPCs) were installed along the perimeter of the port: Tinglado Frutero (TF), Instituto Social de la Marina (ISM), and Dársena Pesquera (DP). These stations act as a barrier between the port and the city, capturing emissions from both port activities and urban sources. To achieve these objectives, 1561 PM10 samples were collected at the TF station, in the port of Alicante, using a High-Volume Sampler (HVS) between 2017 and 2023. In addition, 445 samples were collected at the ISM station between March 2017 and June 2018 using a Low Volume Sampler (LVS). Of the 1241 samples collected with the HVS, corresponding to approximately one sample every two days from 2017 to 2021, 760 filters were subjected to chemical analysis. The chemical analysis of these samples was performed using several techniques to quantify different PM10 components. Ion chromatography (IC, Dionex ICS-1100 and Dionex DX-120) was used to quantify soluble ions, while thermal-optical transmission analysis (TOT, Sunset Laboratory, Inc.) was used to determine carbonaceous compounds. Inductively coupled plasma optical emission spectroscopy (ICP-OES, Optima 7300 DV) and ICP mass spectrometry (ICP-MS, 8900 Triple Quad Agilent) were used to measure major and trace elements. For these analyses, filters were fractionated, and different extraction methods were applied depending on the chemical species studied: a water extraction for soluble ions and a nitric acid (HNO3) + hydrogen peroxide (H2O2) digestion for total metals. Elemental and organic carbon (EC-OC) analyses were performed on filter punches (1,5 cm²), while carbonate concentrations were estimated from ion balance calculations. To ensure the reliability of the data, blank filters were analyzed alongside each batch of samples using the same procedures. The average blank concentration for each species was subtracted from the sample concentrations to account for background contamination. Detection limits were determined as three times the standard deviation of the blank concentrations associated with each filter type. A Positive Matrix Factorization (PMF) analysis was performed using the chemical composition data to identify the sources contributing to PM10 concentrations in the study area. Following the source apportionment, a Mann-Whitney statistical test was applied to compare the factors obtained by PMF and to assess whether different data subsets showed statistically significant differences. Finally, the Theil-Sen estimator was used to analyze long-term trends in PM10 sources over the 2017–2021 period. The first part of the results shows the basic statistics related to the concentrations of PM10 measured in the port of Alicante, such as the temporal evolution of the concentrations, the days of exceeding the maximum limit set by the legislation, maximum and minimum values, etc. from 2017 to 2021. The average annual concentration over the seven years is not statistical different, being 27.5 μg/m³ for the whole period. The maximum daily average concentration was 346.2 μg/m³ in 2022, while the minimum was 1.3 μg/m³ in 2021. The number of exceedances of the daily PM10 limit value (> 50 μg/m³) shows a clear reduction from 2017 to 2018, probably due to the implementation of the abatement measures by the Port of Alicante from 2018 onwards. The number of exceedances shows no statistical difference from one year to another from 2018 to 2023. By removing the Saharan Dust events from the database, a reduction in daily PM10 concentrations ranging from 6 % to 20 % and in the number of daily exceedances o ranging from 33 % to 77 % from 2018 to 2023 was observed. Saharan Dust events have a strong impact on PM10 concentrations in the area, as there is a statistical difference between the dataset including and excluding the Saharan dust events for every year of the study. These differences seem to increase from summer 2020 onwards. In the next two sections, a daily correction is applied to the PM10 automatic data obtained by the OPC located at TF using the High-Volume Sampler of TF. Two datasets are then used, the first one using all the data and the second removing days where the impact of Saharan dust events with concentrations is higher than 2 μg/m³. The Saharan dust events have been identified using meteorological sand dust models and quantified with the use of a regional background station. These datasets are used to study the daily, weekly and monthly evolution of PM10 and NO2. Wind speed is also included to help understand these trends. The analysis of PM10 and NO2 hourly concentrations reveals distinct temporal patterns at different time scales. On a daily scale, the daily cycle of both pollutants showed a peak between 8:00 and 9:00, coinciding with the beginning of the morning traffic activity. At this time, the mixing layer is still shallow due to the lack of solar heating, limiting pollutant dispersion and contributing to the peak. As the day progresses, solar radiation increases, causing the mixing layer to expand and enhancing vertical dispersion. Additionally, wind speeds increase during the early afternoon (reaching their maximum between 14:00 and 15:00), further promoting pollutant dispersion. As a result, concentrations gradually decrease until approximately 18:00. A second rise in PM10 and NO2 levels is observed around 18:00, associated with increased traffic emissions during the evening rush hour and reduced atmospheric dispersion. This behavior is consistently observed over the 7 years of the study. On a weekly scale, the behavior of PM10 and NO2 follows a similar pattern to the daily cycle on weekdays and, to a lesser extent, on Saturdays. However, Sundays show both lower overall concentrations and a less pronounced diurnal variation, probably due to reduced traffic and industrial activity. On weekdays, the hourly PM10 concentrations have a minimum concentration of about 22 to 26 μg/m³, while on weekends this minimum is about 15 y 20 μg/m³. Even though the weekday pattern remains similar, the specific day with the highest concentrations varies from year to year. This variation does not appear to be related to holidays, as the overall weekly distribution remains consistent throughout the study period. These differences are particularly notable in the dataset that includes Saharan dust events. On an annual scale, PM10 concentrations are highest in months with frequent Saharan dust intrusions, especially in February, March, April, August, and September, with monthly average concentrations reaching up to 45 μg/m³, due to very strong Saharan events that occurred in 2020, 2022 and 2023. These intrusions play a significant role in shaping PM10 levels, highlighting the importance of natural sources in the overall concentration trends. After removing the contribution of these events, the monthly PM10 concentrations remain relatively stable and are predominantly influenced by local wind patterns, without a consistent seasonal trend across different years. In contrast, NO2 exhibits a clear seasonal pattern, with higher concentrations recorded between October and March, corresponding to the colder months, and lower concentrations during the warmer period. This trend, observed in every year of the study, is likely to be driven by seasonal variations in atmospheric stability. The wind regime in the study area has remained stable throughout the seven-year study period, showing a consistent alternation between two predominant wind directions: southeast (SE) and northwest (NW). These two directions account for the vast majority of winds affecting the area and define a clear daily cycle. At night and in the early hours of the morning (22:00 to 8:00), low winds blow mainly from the NW (land to sea), facilitating the transport of pollutants generated in urban and industrial areas. In contrast, during the late morning and afternoon hours (11:00 to 20:00), winds shift to the SE (sea to land), favoring the inland transport of maritime and port-related emissions. The transitional periods (9:00 to 10:00 and 21:00 to 22:00) are characterized by the presence of hybrid wind regimes, where the direction is less defined, leading to temporary fluctuations in local air quality conditions. On an annual basis, NW winds are more frequent during the coldest months (October to February), while SE winds predominate during the warmer months (May to August). During transitional months such as March, April, May, and September, both wind directions occur with similar frequency, indicating a more balanced influence of land and sea breezes. Analysis of local PM10 sources using polar plots confirms the presence of three dominant and stable contributors within the study area. Each of these sources is well-defined and has a relatively consistent behavior over time. The first source, originating from the NW, is associated with road traffic, particularly the motorway entrance. This source is a persistent contributor to PM10 levels, as traffic emissions remain a major factor in urban air pollution. The second source, from the SSW, corresponds to Dock-17, a key location for bulk material handling in the port. The impact of this source was significantly reduced during the construction of an enclosed loading and unloading area to reduce emissions. However, when operations resumed in 2023, its impact increased again, highlighting the ongoing challenge of controlling particulate matter emissions from port activities. The effect of this source is particularly pronounced during the summer months, when sea breezes (SE winds) are more frequent, facilitating the transport of port-generated particles towards inland areas. A third source, associated with high PM10 concentrations, has been identified with strong easterly (E) winds, particularly in 2018 and from 2020 onwards. These concentrations appear to be linked to loading and unloading activities at Docks 13 and 11. During the closure of Dock-17 in August 2020, operations were relocated to these docks, which likely contributed to the observed increase in PM10 levels from the east. The chemical characterization of PM10 samples collected between 2017 and 2021 enabled a detailed source apportionment analysis using Positive Matrix Factorization (PMF). This approach successfully identified the main sources contributing to airborne particulate matter in the study area and allowed an assessment of their long-term trends. The identified sources, ranked from highest to lowest impact on PM10 concentrations, are: “Road Traffic”, “Bulk Materials and Dust”, “Aged Sea Salt”, “Saharan Intrusion”, “Fresh Sea Salt”, “Shipping Emissions”, and “Ammonium Sulphate”. These sources were subsequently analyzed to determine their evolution over the five-year period, revealing decreasing trends for "Road Traffic", "Shipping Emissions" and "Ammonium Sulphate" and increasing trends for "Saharan Intrusion" and "Aged Sea Salt". To refine these findings, an additional comparative analysis was conducted between 2017 and 2020, using a more detailed dataset. This analysis allowed for the identification of an eighth source, “Secondary Nitrates”, which had been previously included into the broader categories of "Road Traffic" and "Aged Sea Salt" in the five-year analysis due to the lack of specific chemical tracers in the longer dataset. The decrease in the contribution of the traffic-related source, beyond the temporary reductions observed during the COVID-19 lockdown, appears to be associated with a progressive decrease in the use of private cars until 2022, probably influenced by an increase in teleworking and the modernization of the vehicle fleet towards lower-emission models in line with stricter EURO regulations. Similarly, the reduction in the contribution of the shipping emissions source, which only becomes significant from 2020 onwards, is directly linked to the implementation of IMO2020 regulations, which significantly limited sulfur content of marine fuel oil. The decrease in ammonium sulphate levels can be attributed to several factors, including a reduction in local ammonia emissions, the overall decrease in shipping emissions, and the competitive interactions with alkaline species such as calcium (Ca), magnesium (Mg) and sodium chloride (NaCl), which reduce the availability of sulfate ions (SO₄²⁻) required for ammonium sulfate formation. In contrast, the increase in the contributions from the "Saharan Intrusion" and "Aged Sea Salt" sources over the 2017-2021 period is consistent with long-term meteorological and thermodynamic analyses of African Dust outbreaks. The increase in the “Aged Sea Salt” source is particularly notable and is likely due to an increased frequency of atmospheric recirculation episodes, which prolong the residence time of airborne particles. In addition, a decrease in ammonia (NH3) leads to a higher availability of nitrates and sulfates, further facilitating the formation of aged sea salt particles. Among the sources that did not show significant trends, the contributions from “Fresh Sea Salt” and “Bulk Materials and Dust” sources remained relatively stable over time. Despite the mitigation measures implemented to reduce emissions from bulk materials handling, which have successfully reduced the number of exceedance days associated with this source since 2018, its overall contribution to PM10 concentrations has remained unchanged, suggesting that additional measures may be required to achieve further reductions. The impact of the COVID-19 pandemic on air quality in the study area was also assessed. When comparing annual PM10 concentrations, no statistically significant difference was observed between 2020 and 2017 (the latter being used as a reference year). However, the source apportionment analysis using PMF provided valuable insights into the evolution of emission sources evolved in 2020. The shipping emissions source showed a notable decrease of 77 % over the whole year, with an even more pronounced decrease of 84 % during the lockdown period compared to 2017. This decrease seems to be influenced by both the IMO2020 regulations and the reduction of port activity due to pandemic-related restrictions on maritime traffic. Similarly, the road traffic source showed a 39 % decrease in PM10 concentrations during the lockdown period when compared to the same months in 2017. This decrease correlates with a significant reduction in vehicular circulation in the vicinity of the monitoring site, and a similar decrease in NO₂ concentrations. However, on an annual scale, the contribution of the road traffic remained relatively unchanged, with values of 3,6 μg/m³ in 2017 and 3,7 μg/m³ in 2020. This suggests that despite the temporary decrease in emissions during the lockdown, a subsequent increase in traffic-related emissions for the remainder of the year negating the potential impact that the lockdown could have had on the annual average. Reductions in anthropogenic sources such as traffic and shipping emissions in 2020 were largely compensated by increases in natural and secondary sources, in particular from sources like “Saharan Intrusion”, “Fresh Sea Salt”, and “Aged Sea Salt”. Together, these sources increased from 36% to 51% of total PM10 concentrations between 2017 and 2020, highlighting the complex interplay between local and long-range transport sources. As a result, without the COVID-19-related reductions, PM10 levels in 2020 would have been higher than in the reference year (2017), rather than showing a decrease. The final section of this thesis identifies a problem in the calibration of the Optical Particle Counters (OPCs) in the port area. A comparison between annual gravimetric and OPC PM10 concentrations revealed a systematic underestimation by the OPCs, which increases over the period 2017 to 2023. This growing bias suggests that the calibration applied to the OPCs is no longer adequate and a correction is necessary to ensure consistency over time. To address this, two different correction models of increasing complexity, a multilinear regression model and a Random Forest model, have been developed and tested. These models aim to improve the accuracy of OPC measurements by calibrating them against gravimetric reference data from the TF monitoring station. This station demonstrated a good correlation with OPC measurements due to its proximity to the OPCs, ensuring a reliable reference. In addition to improving measurement accuracy, the corrected OPC data are intended to support the spatial analysis of PM10 variability within the port area, enabling more robust comparisons between locations. The calibration models were initially designed to include five key variables: temperature, relative humidity, time drift, wind speed, and measured OPC uncalibrated concentration. To further assess the influence of wind speed, an extended study introduced additional wind-related variables, including wind speed categorized by quadrant and octant and the daily frequency of winds from each directional sector. The multilinear regression model gave moderate correlation values, with R² ranging from 0.44 to 0.61, depending on the station. Although this model does not achieve the highest accuracy, it already represents a substantial improvement over the raw OPC data. Its main advantage lies in the ability to explicitly quantify the influence of each variable through its coefficients, providing a clearer understanding of the factors influencing OPC measurements. On the other hand, the Random Forest model, although less directly interpretable than the multilinear approach, exploits non-linear relationships within the dataset, leading to a significantly better fit. This results in higher correlation values (R² between 0.67 and 0.80), demonstrating its superior predictive capability. A closer examination of the variable selection process within the Random Forest model provides important insights into the role of wind speed in OPC calibration. In the case of the TF OPC, the model ultimately relies on global wind speed rather than the directionally segmented wind speeds and frequencies. This suggests that, given the proximity of the TF OPC to the TF reference station, the same pollution sources influence both locations, making wind speed primarily a dispersion factor rather than an indicator of source variation. However, for OPCs located further from the reference station (IMS and DP), local PM10 sources vary depending on wind direction. Consequently, the model incorporates specific wind speed and directional components to enhance its predictive performance. This distinction highlights the importance of considering local wind patterns when applying correction models to different monitoring sites. The calibrated data series generated by the Random Forest model show a clear improvement over the original OPC measurements. In particular, the corrected dataset is closer to the TF station reference values than the daily correction dataset used to identify daily, weekly and monthly evolutions previously, further validating the effectiveness of the model. To explore the feasibility of correcting hourly OPC measurements, the Random Forest model trained on daily data was tested on hourly variables. The resulting hourly-corrected dataset was then analyzed to identify local sources contributing to PM10 concentrations at each station. This analysis confirms that the main emission sources affecting all three stations are the activities coming from the port docks, the highway and the main road closest to the port. While the initial application of the Random Forest model to hourly datasets appears promising, the results suggest that training the model directly on hourly data would improve the accuracy and precision of the OPC correction. This step would provide a more refined approach to identifying and quantifying local pollution sources at a finer temporal resolution. The multilinear regression model gave moderate correlation values, with R² ranging from 0.44 to 0.61, depending on the station. Although this model does not achieve the highest accuracy, it already represents a substantial improvement over the raw OPC data. Its main advantage lies in the ability to explicitly quantify the influence of each variable through its coefficients, providing a clearer understanding of the factors influencing OPC measurements. On the other hand, the Random Forest model, although less directly interpretable than the multilinear approach, exploits non-linear relationships within the dataset, leading to a significantly better fit. This results in higher correlation values (R² between 0.67 and 0.80), demonstrating its superior predictive capability. A closer examination of the variable selection process within the Random Forest model provides important insights into the role of wind speed in OPC calibration. In the case of the TF OPC, the model ultimately relies on global wind speed rather than the directionally segmented wind speeds and frequencies. This suggests that, given the proximity of the TF OPC to the TF reference station, the same pollution sources influence both locations, making wind speed primarily a dispersion factor rather than an indicator of source variation. However, for OPCs located further from the reference station (IMS and DP), local PM10 sources vary depending on wind direction. Consequently, the model incorporates specific wind speed and directional components to enhance its predictive performance. This distinction highlights the importance of considering local wind patterns when applying correction models to different monitoring sites. The calibrated data series generated by the Random Forest model show a clear improvement over the original OPC measurements. In particular, the corrected dataset is closer to the TF station reference values than the daily correction dataset used to identify daily, weekly and monthly evolutions previously, further validating the effectiveness of the model. To explore the feasibility of correcting hourly OPC measurements, the Random Forest model trained on daily data was tested on hourly variables. The resulting hourly-corrected dataset was then analyzed to identify local sources contributing to PM10 concentrations at each station. This analysis confirms that the main emission sources affecting all three stations are the activities coming from the port docks, the highway and the main road closest to the port. While the initial application of the Random Forest model to hourly datasets appears promising, the results suggest that training the model directly on hourly data would improve the accuracy and precision of the OPC correction. This step would provide a more refined approach to identifying and quantifying local pollution sources at a finer temporal resolution.