Spatiotemporal Analysis of PM10 and PM2.5 with EBK3D and Space-Time Cube in the City of Lisbon, Portugal

Spatiotemporal Analysis of PM10 and PM2.5 with EBK3D and Space-Time Cube in the City of Lisbon, Portugal João Maria Telo Abreu Jardine Neto ii Spatiotemporal Analysis of PM10 and PM2.5 with EBK3D and Space-Time Cube in the City of Lisbon. Portugal Dissertation supervised by PhD Ana Cristina Marinho da Costa Dissertation co-supervised by PhD Jorge Mateu Dissertation co-supervised by PhD Pedro Cabral February, 2024 iii DECLARATION OF ORIGINALITY I declare that the work described in this document is my own and not from someone else. All the assistance I have received from other people is duly acknowledged and all the sources (published or not published) are referenced. This work has not been previously evaluated or submitted to NOVA Information Management School or elsewhere. Lisbon, February, 26th, 2024 João Maria Telo Abreu Jardine Neto [the signed original has been archived by the NOVA IMS services] iv ACKNOWLEDGMENTS I would like to thank my main supervisor, Prof. Ana Cristina Costa, for her guidance throughout my research. Her support has been essential to my work. My gratitude also goes to my co-supervisors, Prof. Jorge Mateu and Prof. Pedro Cabral, for their valuable input. Special thanks to Nyi Nyi for his assistance and support throughout the thesis process. Thank you to my master's cohort for their camaraderie and support during this journey. Thanks to Leonor Neto for helping me proofreading my manuscript. Finally, I want to thank my family and friends for their support and encouragement. v Spatiotemporal Analysis of PM10 and PM2.5 with EBK3D and Space-Time Cube in the City of Lisbon. Portugal ABSTRACT This thesis conducts a spatiotemporal analysis of particulate matter (PM10 and PM2.5) in Lisbon, Portugal, through 2022, utilizing Empirical Bayesian Kriging 3D (EBK3D) and Space-Time Cube analysis to explore pollution dynamics. Focused on how Particulate Matter (PM) levels vary across Lisbon and identifying distinct patterns during different traffic periods on weekdays and weekends. It employs geostatistical methods to analyze pollution levels, offering insights into the spatial and temporal distribution of PM concentrations. Key findings highlight areas with persistent high pollution and temporal fluctuations throughout the city. This research helps in the understanding of Lisbon's PM related air pollution. Sustainable Development Goals (SDG): vi KEYWORDS Urban Air Quality PM10 PM2.5 Empirical Bayesian Kriging 3D Space-Time Cube Emerging Hot Spot Analysis Local Outlier Analysis vii ACRONYMS ARH - Afternoon Rush Hour EBK3D - Empirical Bayesian Kriging 3D EEA - European Environment Agency ESDA - Exploratory Spatial Data Analysis EU - European Union GIS - Geographical Information Systems IDW - Inverse Distance Weighting MRH - Morning Rush Hour ORH - Off Rush Hour PM – Particulate Matter PM10 - Particulate Matter up to 10 micrometers in size PM2.5 - Particulate Matter up to 2.5 micrometers in size WHO - World Health Organization viii INDEX OF THE TEXT DECLARATION OF ORIGINALITY .......................................................................................................... III ACKNOWLEDGMENTS ........................................................................................................................ IV ABSTRACT........................................................................................................................................... V KEYWORDS ........................................................................................................................................ VI ACRONYMS ...................................................................................................................................... VII INDEX OF THE TEXT .......................................................................................................................... VIII INDEX OF TABLES ................................................................................................................................ X INDEX OF FIGURES ............................................................................................................................. XI 1 INTRODUCTION......................................................................................................................... 1 2 LITERATURE REVIEW ................................................................................................................. 2 2.1 URBAN AIR QUALITY ........................................................................................................................ 2 2.2 PARTICULATE MATTER ..................................................................................................................... 3 2.3 ROAD TRAFFIC RELATED PM.............................................................................................................. 5 2.4 IMPACTS ON HEALTH ........................................................................................................................ 6 2.5 IMPACTS ON ENVIRONMENT .............................................................................................................. 7 2.6 GUIDELINES.................................................................................................................................... 8 2.7 PM DISPERSION .............................................................................................................................. 9 2.8 MITIGATION POLICIES ...................................................................................................................... 9 3 METHODOLOGY ...................................................................................................................... 12 3.1 DATA .......................................................................................................................................... 12 3.2 DATA PREPARATION ...................................................................................................................... 13 3.3 EXPLORATORY SPATIAL DATA ANALYSIS ............................................................................................. 15 3.4 EMPIRICAL BAYESIAN KRIGING 3D .................................................................................................... 15 3.5 SPACE TIME CUBE ......................................................................................................................... 17 3.6 EMERGING HOT SPOT ANALYSIS ....................................................................................................... 18 3.7 LOCAL OUTLIER ANALYSIS ............................................................................................................... 19 4 RESULTS .................................................................................................................................. 20 4.1 EXPLORATORY AND SPATIAL DATA ANALYSIS ...................................................................................... 20 4.1.1 PM10 ............................................................................................................................... 20 4.1.2 PM10 - MRH .................................................................................................................... 20 4.1.3 PM10 - ORH .................................................................................................................... 20 4.1.4 PM10 - ARH ..................................................................................................................... 21 4.1.5 PM2.5 .............................................................................................................................. 22 4.1.6 PM2.5 - MRH ................................................................................................................... 22 4.1.7 PM2.5 – ORH ................................................................................................................... 22 4.1.8 PM2.5 – ARH ................................................................................................................... 23 4.2 EMPIRICAL BAYESIAN KRIGING 3D .................................................................................................... 23 4.3 PM10 EMERGING HOT SPOT .......................................................................................................... 25 4.3.1 PM10 - MRH .................................................................................................................... 25 4.3.2 PM10 - ORH .................................................................................................................... 27 4.3.3 PM10 - ARH ..................................................................................................................... 28 4.4 PM10 LOCAL OUTLIER .................................................................................................................. 30 4.4.1 PM10 - MRH .................................................................................................................... 30 4.4.2 PM10 - ORH .................................................................................................................... 32 4.4.3 PM10 - ARH ..................................................................................................................... 33 4.5 PM2.5 EMERGING HOT SPOT ......................................................................................................... 35 4.5.1 PM2.5 – MRH .................................................................................................................. 35 ix 4.5.2 PM2.5 - ORH ................................................................................................................... 37 4.5.3 PM2.5 ARH ...................................................................................................................... 39 4.6 PM2.5 - LOCAL OUTLIER ............................................................................................................... 41 4.6.1 PM2.5 – MRH .................................................................................................................. 41 4.6.2 PM2.5 – ORH ................................................................................................................... 42 4.6.3 PM2.5 – ARH ................................................................................................................... 43 5 DISCUSSION ............................................................................................................................ 45 6 CONCLUSION .......................................................................................................................... 47 BIBLIOGRAPHIC REFERENCES ........................................................................................................... 49 ANNEX ............................................................................................................................................. 54 APENDIX .......................................................................................................................................... 56 4 vehicle exhaust and power plants), industrial processes, construction and many others. Particles with diameters less than 10 µm can stay in the air for many days and be easily carried by wind, rain or be captured by vegetation or buildings (Roveli,2017 ). Particulate matter formation can be primary or secondary, in which primary particles are realized directly to environment through a certain source (natural or anthropogenic) and secondary sources are formed in the atmosphere as the result of chemical reactions that lead to the creation of particulate matter. The main sources of primary PM in urban areas are: road traffic, fixed combustion (mainly domestic chimneys) and, industrial activities (Guevara,2016). Dust and sea are also important sources of primary PM, and are primarily influenced by wind. Black carbon has a similar chemical structure to graphite, and is mostly generated through improper combustion of fossil fuels (mainly diesel engines) (Zhu, 2002). Examples of secondary PM are sulfates and nitrates which are formed by the oxidation of SO2 and NO2 in the atmosphere into acids (Zheng, 2005). Compared to primary PM, the chemical processes involved in creating secondary PM are relatively slow, however they persist longer in the atmosphere. Particle size is one of the most important characteristics of PM. The way of measuring their size is by defining their aerodynamic diameter, this is defined by the diameter of a spherical particle with a volumetric mass of 1 g.cm-3 (which is the same as water). This method is a simple way of categorize the different sizes of particles with different shapes (Brook, 2010). There are several categories of PM depending on their aerodynamic diameter these are: - Coarse Particles: Between 2.5 and 10 µm - Fine Particles: ≤ 2.5 µm - Ultrafine Particles: ≤ 0.1 µm - Nano Particles: < 100 nm The dimensions of particles not only determine their behavior in the atmosphere but also how they penetrate the human respiratory system. In general, fine particles tend 5 to have easier time penetrating the alveoli (tiny air sacs where the exchange of oxygen and carbon dioxide takes place) and bronchioles (small airways that lead to the alveoli) than coarse particles tend to not pass the nasopharynx (the upper part of the throat that lies behind the nose) (Brooke, 2010). In terms of their composition, PM is made up by multiple components, including black carbon, organic carbon (CO), ion sulfate (SO4), ion nitrate (NO3), metallic composites, material originating from the earth crust, and sea salt. The properties of this complex group of particles can vary according to their composition and size, as well as their impact. For example, black carbon is related with a series of environmental impacts, such as rise in temperature, due to their ability to absorb light. The main components of the material originating from the earth crust include: aluminum (Al), silicon (Si), calcium (Ca), iron (Fe), these are mostly associated with coarse particles. In Europe, this type of particles represents almost 20% of the mass of coarse particles. This is more present in the Southwest and Southeast due to a drier and warmer climate, as well as the influence of dust particles from north Africa (Hajat, 2015). In urban areas road traffic is one of the main sources of PM, followed by fossil fuels combustion in power plants, and industrial manufacturing (Lelieveld,2015). Besides engine combustion (which can vary depending on the age of the engine and the type of fuel used), there are other sources of PM related to road traffic, such as, brake pads wear, tire wear, and re-suspension of dust in the road surface (Bond, 2013). 2.3 Road Traffic related PM Multiple studies have shown that the majority of emissions of PM in urban areas are related to road traffic. (Pant,2013; Ying,2015; Chih-I,2023) The PM emissions related to road traffic are classified according to their formation process. Internal combustion engines, both gasoline and diesel are pointed as the main mechanism to which PM is formed in urban areas. Additionally, road traffic related activities can also be the source of PM (Bond,2013). Comparing exhaust emissions of motor vehicles running on gasoline and diesel shows that diesel engines release more PM into the atmosphere, compared to gasoline engines, studies also show that diesel heavy commercial vehicles are the ones that 6 produce more PM in the diesel category. Analyzing non exhaust related PM in cars can be difficult, as they can be influenced by many different factors, such as, the properties of the materials (type o tire, type of break, how rough the pavement is), and weather factors (temperature and humidity). Different countries or manufacturers can have multiple standards of construction, and for that reason analyzing non exhaust related PM emissions across Europe can show considerable variability (Thorpe,2008 ). 2.4 Impacts on health Prolonged exposure to PM can have adverse health effects on human health and can be related to different health conditions, such as cardiovascular diseases, respiratory diseases, cancer, pregnancy complications and may others (Lee, 2021). Assessing the adverse effects of PM can be difficult because the air composition can vary from place to place and individuals can be exposed to a vast mixture of substances, that can have different effects depending on the person and the time of exposure (Bastos, 2019). Operators of motor vehicles, passengers, pedestrians, or cyclists are often exposed to high concentrations of PM due to the proximity of roads in normal day to day activity. Operators and passengers of motorized vehicles have a higher probability of exposure to PM than does that travel by foot or cycle. However, considering that the ladder is directly exposed to the air (not inside a motor vehicle) as well as longer commute times can lead to higher doses of PM inhaled and deposited in the respiratory system compared to does who travel in a motor vehicle (Bastos, 2019). There are multiple scientific studies that demonstrate the relationship between exposure to PM and adverse health effects, independent of the level of development of the country. WHO establishes a clear relation between multiple and diseases and even premature deaths related to PM, mainly PM10 and PM2.5(WHO, 2018). PM2.5 in particular, due to its size, can easily penetrate the human respiratory system, and can even reach areas outside the pulmonary system, such as the nervous system. Evidence also suggests that PM can cause irritation in the pulmonary system and escalate existing chronic pulmonary diseases, such as asthma. PM is also composed of 7 different components which are harmful for the health, such as heavy metals, carbon composites, and even cancerogenic elements (WHO, 2018). Studies also reported that short and long-term exposure to PM2.5 can be associated with multiple health risks, such as cancer. In 2013 the IARC (International Agency for Research on Cancer) added PM2.5 to its list of cancerogenic substances for human beings (IARC, 2013). Coarse particles (PM10-2.5) ignore the natural defenses of the body like the nose and throat depositing themselves on the thorax. These can be expelled by the body through coughing or sneezing, however as PM2.5 due to its size, are capable of penetrating deeper into the lungs, and can cause problems in the lungs and even hearth. Moreover, they can introduce harmful substances into the bloodstream and can stay in the body for longer periods of time (Chair, 2010). 2.5 Impacts on Environment PM can have different impacts on the environment such as reduced visibility, mainly a problem in urban areas. The level of impact on the visibility depends based on multiple factors, such as density of the concentrated particles, the size of the particles, weather conditions, time of day, etc. Related to the lack o visibility is the reduction of solar radiation caused by the absorption of solar rays by the particles. This can lead to a decrease in temperature which can lead to serious environmental impacts (Dubey, 2018) PM can also cause harm in vegetation and water bodies, if its concentrations are high and toxic or of big dimensions, they can reduce crop production blocking plants natural process. If they are alkaline or acid they can change the pH of water bodies, and if the particles are toxic they can get into contact with plants and animals and affect the food chain (Dubey,2018). 8 2.6 Guidelines Despite efforts to improve air quality within the EU, many countries still registered exceed the quality standards established by the 2008 Air Quality Directive, an agreement among all member states setting pollutant threshold limits in the EU. These legally binding thresholds, transposed into national law for each member state, for Portugal thresholds are under the Directive 2008/50/CE of May 21. These values are notably less strident than those recommended by the World Health Organization (WHO). Table 1 delineates the standards defined by EU and WHO. Directive 2008/50/CE of May 21 WHO Daily Limit Thresholds (24h) (average) Annual Limit Thresholds (average) Daily Limit Thresholds (24h) (average) Annual Limit Thresholds (average) PM2.5 - 25 µg.µm³ 25 µg.µm³ 10 µg.µm³ PM10 50 µg.µm³, not exceed more than 35 days in one year 40 µg.µm³ 50 µg.µm³ 20 µg.µm³ Table 1: EU and WHO Thresholds As reported by (Lopes, 2019), the European Environmental Agency’s Air Quality Report (EEA,2018) reveals that 19% of air quality monitoring stations recorded exceedances of daily PM10 limit. Furthermore, annual PM10 concentrations breaches were detected at 6% of these stations. Regarding PM2.5 levels, annual limits were exceeded at 5% of stations throughout the EU. When assessed against WHO’s stricter PM10 guidelines, 48% of stations from al reporting countries (with the exceptions of Estonia, Iceland, Ireland, and Switzerland) exceeded these thresholds. Similarly, for PM2.5, 68% of stations from the reporting countries (excluding Estonia, Finland, Hungary, Norway, and Switzerland) exceeded WHO’ recommended limits. This report also highlights the stark contrast between EU and WHO standards, showing that in 2016, 13% of urban areas in the EU experienced levels surpassing EU limits for pollutants, a figure that escalates to 42% under WHO guidelines. For PM2.5 exposure, only 6% of the EU’s urban population were subjected to concentrations above EU limits, a percentage that dramatically increases to 74% when evaluated against WHO standards. 9 The challenge of aligning with WHO’s PM guidelines remains significant for Europe. Despite initiatives to reduce air pollution, numerous EU nations still report PM concentrations levels above the thresholds establishes by the 2008 directive. This gap between the EU and WHO standards highlights the urgent need for enhanced and more stringent measures to achieve the 2030 objective of meeting WHO guidelines. 2.7 PM dispersion Weather conditions heavily affect PM dispersion being composed of two main components: - Vertical component created by the turbulence generated by the thermic vertical gradient between the lower levels of the atmosphere. - Horizontal component in which the wind is the main component in both transport and mixture. Atmospheric processes and the circulation of high-pressure centers determine the weather above continents and oceans. High pressure centers also known as anticyclones are associated with great stability and low vertical mixture, for that reason, there is low dispersion of PM. Low pressure centers are associated with conditions of atmospheric instability with great turbulence which favors PM dispersion. In winter colder days, the sun during the day heats the air next to the surface of earth. At the end of the day, after the sun sets there is a quick cooling of the air which is mostly felt closer to the ground. When lower layers of the atmosphere have lower temperatures than the ones at higher altitudes, it occurs a thermic inversion, which creates high atmospheric stability, and the dispersion conditions are lower. For this reason concentrations measured during this time period might be higher than the others (Zheng, 2005). 2.8 Mitigation Policies Due to the increase in emissions produced by motorized vehicles, in particular urban areas, there was an increase concern by government bodies to control this type of 10 pollution. As mentioned above the EU directive of 2008 set goals and measures for member states to implement in order to improve Europe´s air quality. The concept of sustainable mobility is widely used in this document and is one o the main strategies to mitigate pollution levels. The goal is to increase the energy efficiency of movement in order to reduce its environmental impact. The main objectives of sustainable mobility are as follows: - Minimize the use of personal transportation. - Optimize the use of public transportation. - Increase the use of soft modes of transportation (these are ways of transportation more friendly for the environment such as walking or cycling) The implementation of these policies requires deep urban changes not only in a physical sense, with changes to urban infrastructure but also how the public sees these policies and the advantages of adopting to more environmentally friendly modes of transportation. There needs to be changes to the structure of territories, public transport infrastructure and public places better design to adopt these changes. One of the first strategies implemented by member states to mitigate this type of pollution are the Low Emissions Zones (LEZ), these zones are designed to limit or restricted access to certain urban areas by certain types of vehicles. In order to study the capability of LEZs a study was conducted in which analyzed changes in air quality of five EU member states that implemented LEZ (Denmark, Germany, Netherlands, Italy and UK). (Guevara, 2016) The study showed mixed results, although Germany showed a reduction in annual PM10 concentrations of up tp 7% in those areas, the same results weren’t shown in different urban areas. These results might be related to the type of vehicles that were restricted in German ZER (it restricts the entry to diesel vehicles and passenger vehicles while other countries only restricted diesel vehicles). The low percentage of the reduction of PM10 might also be due to the fact that the ZER limitations dint cause an impact in PM10 emissions of other sources (non-vehicular exhaust emissions), which represent a significant percentage of primary emissions of road traffic. The Implementation of LEZs is not the only project used to better air quality, there are other measures of local, national and international planning and transport policies 11 being implemented to help mitigate road traffic emission in urban areas. In New York City taxing fees were used in areas with high road traffic to increase the cost of circulation in certain areas during certain periods of the day. The goal was direct traffic flow to less congestion roads across the city. (Schaller, 2010) In Beijing, China a temporary restriction of driving based on license plate numbers was used to control the number of cars allowed to drive every day. The system was based on the last digit of the license plate there was a day for odd and even numbers throughout the week. This system was used during the 2008 Olympic games and showed promising results in the reduction of road traffic emissions. After the games were held a less restrictive system was implemented and the restriction happen only once a week. However, it was slowly bypassed by the population as people bought both an even and odd license plate (Huijuan, Fujii & Managi, 2013). 12 3 METHODOLOGY This chapter describes the methodology used in this thesis, which was divided into six stages, as seen on Figure 1. The first stage of data preparation involved organizing the collected data, addressing any missing, incorrect, or inconsistent values to ensure data integrity. The next stage is the Exploratory Spatial Data Analysis (ESDA), where data was characterized using both spatial and descriptive statistics. The third stage involved doing a 3D interpolation of the data using the Empirical Bayesian Kriging 3D (EBK3D) method. The fourth stage involved taking the output of the EBK3D interpolation and integrating it in a Space-Time Cube using ArcGIS Pro software. The fifth and six stages an Emerging Hot Spot Analysis and Local Outlier Analysis was made using the Space-Time Cube created in the fourth stage. 3.1 Data The data used in this thesis is from an air quality monitoring network that is provided by the Lisbon municipality under the “Lisboa Aberta” program (Lisboa Aberta, 2024). The stations are managed by the consortium of MEO/Monitar/QART, and is also part of the European union program “Sharing Cities”. This network has been operational since august 2021, and is comprised of 80 monitoring stations that monitor different Figure 1: Methodology Flow Chart 13 environmental parameters, including PM10 and PM2.5 (Figure 2). These stations are equipped with sensors that measure the hourly average concentrations of these two types of PM. The location of each sensor may vary between light post or building. This monitoring network serves to reenforce the previous fixed monitoring network which at the time was only composed of 6 stations in the city of Lisbon. Due to these stations only been online since august 2021, for this study year 2022 was selected as it is at the time of writing this thesis the only year to which there is full data. There is no metadata available related to the type of stations (if they are related to traffic or background), or the exact positioning of the stations (where they are mounted, at what height, etc.) which can influence the data as buildings play a crucial role in PM dispersion, so this needs to be taken into account when analyzing the results. Figure 2: Monitoring Stations 3.2 Data Preparation In this chapter the data preparation process will be explained. The initial step involved gathering PM10 and PM2.5 concentrations data for the year 2022 (as mentioned before it was the only available year) from all 80 monitoring stations. The collected 20 4 RESULTS 4.1 Exploratory and Spatial Data Analysis 4.1.1 PM10 4.1.2 PM10 - MRH The results of the ESDA for PM10 MRH on week days and weekends, are described in Appendix A1.1 and A1.2 respectively. For week days the mean and standard deviation were greater on the months of January and February, and October through December with December having the highest values. The lowest values were registered in the months of April through June, overall, the mean values of every month stayed below the limit threshold, although in every month some stations had values over the limit threshold which is evident on the maximum values and the IDW, mainly on the months of October through December. Values were positively skewed, except for February, March, and October. Kurtosis values were for January and February, indicating that the distribution had heavier tails and more extreme values than the normal distribution. For the weekends mean and standard deviation values were lower than the week days but with a similar pattern. Only the months of May and August had negative skewness, with the rest being positive. Kurtosis was lower than 3 in January, February, August, and November, with the remaining months being higher. June and December had much more extreme value than the rest of the months. Overall values above the limit threshold were scientifically lower on the weekends than the week days for the same time period. 4.1.3 PM10 - ORH ORH for week days and weekends is described in Appendix A1.3 and A1.4 respectively. For week days the mean values were higher on the months of January, February, and September through December. The lowest mean values were between April and June with a slight increase in July and August. Standard deviation was roughly the same, except for September which was much higher than the others. In terms of skewness all months had positive values. Kurtosis was lower than 3 in January, February, April, August, November, and December. September had the 21 highest value due to having very extreme value compared with the normal distribution. For the weekends, similar to the previous time period the mean and standard deviation values are lower on the weekends compared to week days, with similar trend regrading each month. All months showed positive skewness. Kurtosis was higher than 3, except for January and February, with the highest values in July and December. Overall ORH week days and weekends show similar patterns compared to the MRH, although ORH mean values are lower than MRH on both periods, ORH weekends showed a bigger number of exceeding levels of PM10 than MRH on the same days. 4.1.4 PM10 - ARH ARH is described in Appendix A1.5 for weekdays and A1.6 for weekends. For week days mean and standard deviation values were higher in January, February, August, October, November and December, with December having much higher values than the rest of the months. The lowest values were in the months of April through June. For this time period August had bigger values compared to the trend of the other periods analyzed. All months were positively skewed. Kurtosis was higher then 3, except for the months of January, February, October, and November. August had much higher kurtosis value than the others, due to having higher and more extreme values compared with the normal distribution. For the weekends mean and standard deviation values were much lower than the weekdays, similar to the pattern observed before with the exception of September which had much lower values, and December with much greater values compared with the rest of the months. The skewness values were positive, except for August. Kurtosis was higher than 3, except for January, indicating that the distribution had more extreme values than the normal distribution. Overall, ARH on week days is slightly lower than the MRH on most months except for December, and slightly bigger than ORH. For weekends ARH is generally the lowest of the three except for the month of December. 22 4.1.5 PM2.5 4.1.6 PM2.5 - MRH MRH is described in Appendix A2.1 and A2.2 for week days and weekends respectively. For weekdays the mean and standard deviation were greater in January, February, and September through December, with December having the highest value very close of passing the limit threshold, but overall, all months were below the limit. The skewness was positive across all months. Kurtosis was lower than 3, except for September, meaning that the distribution had lighter tails and less extreme values than the normal distribution. For the weekends standard deviation was stable across all months and the mean was also very similar, June had the lowest mean value and December the highest, however the difference is not that significant. All months had a positive skewness. Kurtosis was higher than 3, except for February, April, and September, which indicates the presence of more extreme values than the normal distribution. Overall MRH on week days is higher when compared with other time periods across almost every month. On weekends MRH is higher than other time periods in March, April, May, July, and August. 4.1.7 PM2.5 – ORH ORH week days and weekends, is described in Appendix A2.3 and A2.4, respectively. On week days the standard deviation is stable across every month. Mean values are highest between January, February March, and October through December. All months are positively skewed. Kurtosis values are lower than 3, except for December, which means the distribution had lighter tails and fewer extreme values compared with the normal distribution. On weekends the standard deviation is stable across every month as well as the mean, with the highest value being December but overall the mean trend remains the same as other time periods with week days having higher mean values than weekends. Skewness values were all positive. Kurtosis values higher then 3, except for April and May, similar to MRH weekend values tend to have a higher kurtosis. Comparing ORH with other time periods, we can say that ORH is the lowest of the three on both week days and weekends. 23 4.1.8 PM2.5 – ARH ARH on week days and weekends is described in Appendix A2.5 and A2.6. For the week days mean values are below the limit threshold with December coming very close to exceeding, values the lowest between April and July. Standard deviation values are stable and positively skewed across all months. Kurtosis is lower than 3, except for the month of May and September, similar to the other two time periods where on week days the kurtosis is lower than three with some exceptions. For weekends the mean and standard deviation are higher in January and December, with a overall trend of the mean similar to other time periods. Skewness is positive across all months. Kurtosis is higher than 3 in the months of February, March, May, August, September, and October, indicating higher or more extreme values in these months compared to the others. Overall comparing ARH is very similar with the MRH period slightly lower on average but shows very similar results on week days. During the weekends tends to be higher than MRH period during the months of January, October, November, and December. 4.2 Empirical Bayesian Kriging 3D EBK3D cross-validation statistics can be seen in Table 2. Overall, the method performed reliably with minor changes to the default advanced model parameters. There were some differences in the performance of the two different PM, but both achieved good results and had minor differences of results for each of the time periods. Different search neighborhoods were used for PM10 and PM2.5 mainly based on the visual aspect of the model and not based on the cross-validation results which showed insignificant changes. Using different sector types mainly impacted the visual aspect of the interpolated surface, and were selected based on these criteria alone. Looking at Table 2 the performance of each model was slightly different on each case, when setting the model parameters and analyzing the cross-validation results minor changes could be made to each, although they weren’t significant and for that reason it was better to use a similar model for each case, changing the sector and neighborhood if significant changes were visualized. On most cases the ME and SME were both close to 0, showing a unbiased prediction. ASE was close to the RMSE and the RMSSE was close to 1 for PM10 weekends MRH, ORH, and ARH, week days didn’t show as good results but it was still very close to that target, PM2.5 24 performed the best on both week days and weekends, with the exception of week days ARH. The percentage points inside the 95 percent interval were, with few exceptions, close to 95% suggesting that the model had and overall good performance. Average CRPS results showed slightly worse results for PM10 week days on the three periods and for PM2.5 weekends ORH. Analyzing the results, we can observe that generally the number of samples had a slight impact on the model performance, although the differences in the number of samples for each model is not significant. MRH ORH ARH PM10 Week Days Average CRPS 1,9253186 2,954744 0,001139868 Inside 90 Percent Interval 93,628088 94,72991 92,53112033 Inside 95 Percent Interval 95,968791 96,31094 95,02074689 Mean -0,345395 -0,56507 -0,193787353 Root-Mean-Square 1,6717398 1,707321 2,730620898 Mean Standardized -0,034116 -0,06365 -0,014231785 Root-Mean-Square Standardized 0,5605864 0,976796 1,10657277 Average Standard Error 0,3723336 0,171405 0,121982943 Weekends Average CRPS 0,377323 1,634583 0,678411438 Inside 90 Percent Interval 91,961853 93,21383 91,25326371 Inside 95 Percent Interval 95,776567 96,15877 96,08355091 Mean -0,064457 -0,10534 -0,182743435 Root-Mean-Square 0,0265319 0,738225 0,674823928 Mean Standardized -0,003361 -0,01519 -0,016973458 Root-Mean-Square Standardized 0,6375992 0,774436 1,162878174 Average Standard Error 0,0350026 0,116884 0,995211389 PM2.5 Week Days Average CRPS 0,6879175 2,066574 0,05082201 Inside 90 Percent Interval 91,454082 91,21172 91,28686327 Inside 95 Percent Interval 95,642857 94,94008 95,30831099 Mean -0,088983 -0,15541 -0,232171836 Root-Mean-Square 0,7507313 0,218055 1,273589524 Mean Standardized -0,029168 -0,02872 -0,050844966 Root-Mean-Square Standardized 0,2295961 1,059213 0,128480813 Average Standard Error 0,3556574 1,262782 0,23568247 Weekends Average CRPS 1,0612272 0,981601 0,234333473 Inside 90 Percent Interval 90,087829 89,94845 90,33942559 Inside 95 Percent Interval 95,483061 94,71649 95,03916449 Mean -0,147496 -0,1545 -0,081631622 Root-Mean-Square 0,9669044 0,07903 0,506472409 Mean Standardized -0,039727 -0,03429 -0,018424825 Root-Mean-Square Standardized 0,0813832 0,015033 0,103117343 Average Standard Error 0,970639 0,424993 0,493957064 Table 2: EBK3D Model Parameters 25 4.3 PM10 Emerging Hot Spot 4.3.1 PM10 - MRH Based on the space-time cube of PM10 MRH (Figure 4), several hot and cold spots were identified (Table 3). For week days 10 types of spatial-temporal cold and Hot Spots were identified. Hot Spots were distributed mainly in the middle north and south of Lisbon creating a corridor of Hot Spots, while cold spots were identified in the east and west of the city. The categories Hot Spots include: consecutive, intensifying, new, oscillating, persistent, and sporadic. The number of Intensifying Hot Spots was the largest, meaning that these locations tend to increase over time and that its increase is statistically significant (Esri, 2022d). Persistent Hot Spots were the second highest, concentrated mainly in the north of Lisbon in the parishes of Santa Clara, Lumiar, part of Olivais (around the Lisbon Airport), and also near Penha de França, which indicates that these areas persistently a hot spot but statistically stable through time. Consecutive Hot Spots appear to border the Intensifying Hot Spots and represent a location generally aren’t statistically significant Hot Spots until the last three time-step intervals (October, November and December). The areas that were identified as cold spots are mainly located in the western regions of Lisbon such as Belém, Ajuda, and Benfica, as well in the east around Marvila. The types of cold spots included: diminishing, Intensifying, persistent. Diminishing cold spots were the largest, meaning those areas are always statistically significant cold spots and keep decreasing through time. Persistent and Intensifying Cold Spots identify areas that are always statistically significant cold spots or were otherwise not significant until the last time-step interval. Regarding the weekends during the MRH, the overall region of cold and Hot Spots remains the same, however the type of hotspot changes, with oscillating hot spot being the largest, meaning that has been a statistically significant hot spot in less than 90 percent of the time and oscillating between the two. Sporadic hot spot is the second largest which is similar to the oscillating in that is typically and on-again off-again hot spot. 26 Overall, the main differences between MRH week days and weekends, is that the hot and cold spots tend to vary in time more on the weekends than the weekdays, meaning that on the week days the pattern tends to be more consistent through time. Figure 4: PM10 Morning Rush Hour, Emerging Hot Spot Analysis WEEK DAYS COUNT % WEEKENDS COUNT % CONSECUTIVE HOT SPOT 42 7,29 Consecutive Hot Spot 26 4,52 DIMINISHING COLD SPOT 77 13,37 Diminishing Cold Spot 39 6,78 INTENSIFYING COLD SPOT 60 10,42 Historical Cold Spot 12 2,09 INTENSIFYING HOT SPOT 105 18,23 Intensifying Cold Spot 35 6,09 NEW HOT SPOT 12 2,08 New Hot Spot 18 3,13 NO PATTERN DETECTED 119 20,66 No Pattern Detected 160 27,83 OSCILLATING HOT SPOT 1 0,17 Oscillating Hot Spot 99 17,22 PERSISTENT COLD SPOT 59 10,24 Persistent Cold Spot 61 10,61 PERSISTENT HOT SPOT 81 14,06 Persistent Hot Spot 49 8,52 SPORADIC HOT SPOT 20 3,47 Sporadic Cold Spot 3 0,52 SUM 576 100% Sporadic Hot Spot 73 12,70 Sum 576 100% Table 3: PM10 Morning Rush Hour, Emerging Hot Spot Analysis Classes 27 4.3.2 PM10 - ORH The results of the emerging hot spot analysis for ORH can be found in Figure 5, as well as Table 4 with the counts of each class. During the weekdays, the results show different types of hot and cold spots. A significant number of Intensifying Hot Spots (26,7%) were observed, particularly in the south-central parishes of Santa Maria Maior, São Vicente, Penha de França, Arroios, as well as the center parts of Lisbon such as Avenidas Novas, parts of Alvalade, and at the north in Lumiar and Santa Clara, indicating a rise in PM10 levels over time. Persistent Hot Spots concentrated mainly in the parish of Misericordia, reflect areas with stable but elevated PM10 values across time. Regarding the cold spots, persistent cold spots were predominantly found in the western regions of the city, in areas such as Belém, Ajuda, Alcântara, and Benfica, suggesting these areas have consistently lower PM10 values. Figure 5: PM10 Off Rush Hour, Emerging Hot Spot Analysis On the weekends, PM10 patterns shift slightly. The most prominent spots are Intensifying Cold Spots in the parishes of Belém and Ajuda, indicating a growing 28 trend towards lower values. Intensifying Hot Spots are observed in Misericordia, Santa Maria Maior, and Santo Antonio, suggesting these areas have an increase in PM10 during the weekends. Additionally, persistent Hot Spots in Lumiar and Santa Clara, and persistent cold spots on the western edge of Ajuda and parts of Benfica indicate areas where PM10 values remain consistently high or low through time, throughout the weekends. Sporadic Hot Spots in Avenidas Novas represent locations with fluctuating PM10 values, typically not consistent enough to form a pattern. Overall, the specific types of hot and cold spots may vary between week days and weekends, the regions of Lisbon experiencing these phenomena remain consistent, with central areas more prone to Hot Spots and peripheral areas to cold spots. 4.3.3 PM10 - ARH Emerging hot spot analysis results for ARH on week days and weekends can found in Figure 6 and Table 5. During the week days, Intensifying Hot Spots are prevalent, especially in the northern parishes of Lumiar, Olivais, Santa Clara, and southern areas such as Santa Maria Maior, São Vicente, and Santo António where PM10 levels tend to rise. The persistent Hot Spots, are located between Arroios, Santo António, and São WEEK DAYS COUNT % WEEKENDS COUNT % CONSECUTIVE HOT SPOT 13 2,26 Consecutive Cold Spot 2 0,35 DIMINISHING COLD SPOT 34 5,90 Consecutive Hot Spot 5 0,87 HISTORICAL COLD SPOT 1 0,17 Diminishing Cold Spot 45 7,81 INTENSIFYING COLD SPOT 28 4,86 Diminishing Hot Spot 1 0,17 INTENSIFYING HOT SPOT 154 26,74 Historical Cold Spot 2 0,35 NEW HOT SPOT 5 0,87 Intensifying Cold Spot 47 8,16 NO PATTERN DETECTED 131 22,74 Intensifying Hot Spot 95 16,49 PERSISTENT COLD SPOT 140 24,31 New Hot Spot 9 1,56 PERSISTENT HOT SPOT 57 9,90 No Pattern Detected 112 19,44 SPORADIC COLD SPOT 2 0,35 Persistent Cold Spot 113 19,62 SPORADIC HOT SPOT 11 1,91 Persistent Hot Spot 111 19,27 SUM 576 100% Sporadic Cold Spot 3 0,52 Sporadic Hot Spot 31 5,38 Sum 576 100% Table 4: PM10 Off Rush Hour, Emerging Hot Spot Analysis Classes 29 Vicente signifying locations that are always high compared to its neighbors. Sporadic Hot Spots are scattered across Avenidas Novas and areas bordering the Intensifying Hot Spots, indicating areas with variable but occasionally high PM10 values. Persistent cold spots make up a significant portion of the results, particularly in the western regions such as Belém, Ajuda, and south of Benfica, indicating consistently lower PM10 values. Diminishing cold spots located in the east in Marvila and the west in Alcântara, and Benfica show areas where PM10 values are decreasing in time. Figure 6: PM10 Afternoon Rush Hour, Emerging Hot Spot Analysis On the weekends, the pattern shifts slightly. The largest category of Hot Spots is the sporadic Hot Spots, indicating that central areas like Penha de França, São Vicente, Santa Maria Maior, Areeiro, Avenidas Novas, and Alvalade experience variable PM10 values through time, and are occasionally statistically significant Hot Spots. Persistent Hot Spots are more prominent in the areas of Arroios, Santo Antonio, Lumiar, and near the airport indicating that these are always significantly statistical hotspots. Historical cold spots in Campo de Ourique and near Campolide, as well as, persistent cold spots in Ajuda and Benfica indicate areas where lower PM10 values are a trend in time. 36 Figure 10: PM2.5 Morning Rush Hour, Emerging Hot Spot Analysis Weekends, persistent cold spots are very similar in terms of areas than the week days. Persistent Hot Spots emerge in the weekends in São Vicente, Penha de França, and Arroios, where high values have been consistently observed to tend to increase over time. Central and northern regions experience Intensifying Hot Spots but at a lower number than the week days, indicating a high but stable trend through time. No pattern detected remains similar to the week days. Comparing MRH period across weekdays and weekends, it is clear that certain areas exhibit persistent trends, either hot or cold, indicating stable values throughout the year. Week days Count % Weekends Count % Consecutive Cold Spot 3 0,520833 Consecutive Cold Spot 10 1,736111 Consecutive Hot Spot 3 0,520833 Consecutive Hot Spot 2 0,347222 Diminishing Cold Spot 18 3,125 Diminishing Cold Spot 15 2,604167 Diminishing Hot Spot 11 1,909722 Diminishing Hot Spot 12 3,819444 Historical Cold Spot 4 0,694444 Historical Cold Spot 2 0,347222 Intensifying Cold Spot 42 7,291667 Historical Hot Spot 1 0,173611 Intensifying Hot Spot 93 16,14583 Intensifying Cold Spot 31 5,381944 37 4.5.2 PM2.5 - ORH Emerging hot spot analysis of PM2.5 ORH are presented in Figure 11 and Table 10. During the week days the persistent cold spots are largely situated in Alcântara, Belém, Benfica, and Carnide. These locations consistently exhibit lower PM2.5 values compared with its spatiotemporal neighbors, Intensifying Hot Spots (18,06%), emerge strongly in the central and downtown parishes such as Beato, São Vicente, Penha de França, Santa Maria, Misericordia, Arroios, Avenidas Novas, and extend to the north areas like Lumiar and Olivais north near the Lisbon Airport, indicating not only persistently high but also growing levels compared with its neighbors in time. Diminishing cold spots (3,82%) are present in locations such as Benfica and Alcântara, and show low values that are diminishing over time compared to its surroundings. No pattern detected (24,31%), split the study area in two between cold and Hot Spots, indicating region where no significant trend has been identified. New Hot Spot 3 0,520833 Intensifying Hot Spot 27 4,6875 No Pattern Detected 159 27,60417 New Cold Spot 2 0,347222 Persistent Cold Spot 165 28,64583 New Hot Spot 7 1,215278 Persistent Hot Spot 62 10,76389 No Pattern Detected 110 19,09722 Sporadic Hot Spot 13 2,256944 Persistent Cold Spot 172 29,86111 Sum 576 100% Persistent Hot Spot 145 25,17361 Sporadic Cold Spot 9 1,5625 Sporadic Hot Spot 21 3,645833 Sum 576 100% Table 9: PM2.5 Morning Rush Hour, Emerging Hot Spot Analysis Classes 38 Figure 11: PM2.5 Off Rush Hour, Emerging Hot Spot Analysis On the weekends the pattern adjusts slightly, with persistent cold spots now accounting for 37,50%, in the same areas as week days but with additional locations in the central areas. Intensifying Hot Spots (14,58%) are present in the north in the parishes of Lumiar, Santa Clara, and Olivais, as well as south areas, indicating a trend of increasing levels through time in relation to its spatiotemporal neighbors. No pattern detected (22,22%) is present in areas such as Parque das Nações and certain central regions that typically exhibit some pattern. Diminishing cold spots (4,69%) and sporadic Hot Spots (2,43%) are still present but in smaller percentages suggesting localized and less consistent patterns during the weekends. Comparing the weekdays to weekdays the most noticeable difference between the two is that during the weekends cold spots tend to take over a larger area of the city while Hot Spots are denser with a trend of increasing through time. Week days Count % Weekends Count % Consecutive Cold Spot 3 0,52 Consecutive Cold Spot 8 1,39 Consecutive Hot Spot 10 1,74 Diminishing Cold Spot 27 4,69 Diminishing Cold Spot 22 3,82 Historical Cold Spot 1 0,17 Historical Cold Spot 1 0,17 Intensifying Cold Spot 3 0,52 39 4.5.3 PM2.5 ARH During week days Figure 12 and Table 11 persistent cold spots (30,56%) are predominantly located in the western regions of Alcântara, Ajuda, belém, and Carnide. Figure 12: PM2.5 Afternoon Rush Hour, Emerging Hot Spot Analysis Intensifying Cold Spot 19 3,30 Intensifying Hot Spot 84 14,58 Intensifying Hot Spot 104 18,06 New Cold Spot 1 0,17 New Hot Spot 4 0,69 New Hot Spot 2 0,35 No Pattern Detected 140 24,31 No Pattern Detected 128 22,22 Persistent Cold Spot 177 30,73 Oscillating Hot Spot 1 0,17 Persistent Hot Spot 78 13,54 Persistent Cold Spot 216 37,50 Sporadic Cold Spot 2 0,35 Persistent Hot Spot 78 13,54 Sporadic Hot Spot 16 2,78 Sporadic Cold Spot 13 2,26 Sum 576 100% Sporadic Hot Spot 14 2,43 Sum 576 100% Table 10: PM2.5 Off Rush Hour, Emerging Hot Spot Analysis Classes 40 These areas consistently show lower PM2.5 levels, than its space time neighbors. Persistent Hot Spots (23,44%), are concentrated in the central areas such as Misericordia, Santa Maria Maior, São Vicente, Penha de França, Arroios, Santo António, Avenidas Novas, Campolide, and in the north areas of Lumiar, Santa Clara, and Olivais, indicating areas where PM2.5 are high but stable over time. No Pattern Dectected (32,64%), includes areas like Carnide, Alvalade, and Olivais, indicating areas where values don’t exhibit any pattern through space and time. On the weekends, the pattern of cold and hotspots shifts slightly. Persistent cold spots are substantial (32,81%), with a distribution similar to week days but extending to the western areas of Marvila, Parque da Nações, Olivais. Intensifying Hot Spots are in are larger the persistent Hot Spots and are located in the north, south areas as well as Belém, which during the week days shows intensifying or persistent cold spots. No pattern detected (28,99%) increases on the weekends with a bigger change being in the west parish of Belém and Ajuda. Comparing the two, the main difference is the shift from persistent Hot Spots during week days to Intensifying Hot Spots on the weekends. This transition indicates that Hot Spots during the week tend to be more stable during time while on the weekends, they tend to increase through time. Week days Count % Weekends Count % Consecutive Cold Spot 1 0,17 Consecutive Cold Spot 15 2,60 Diminishing Cold Spot 7 1,22 Consecutive Hot Spot 8 1,16 Diminishing Hot Spot 11 1,91 Diminishing Cold Spot 2 0,35 Historical Cold Spot 3 0,52 Diminishing Hot Spot 1 0,17 Intensifying Cold Spot 15 2,60 Historical Cold Spot 2 0,35 Intensifying Hot Spot 17 3,65 Intensifying Cold Spot 6 1,04 New Hot Spot 3 0,52 Intensifying Hot Spot 101 17,53 No Pattern Detected 188 32,64 New Hot Spot 6 1,04 Persistent Cold Spot 176 30,56 No Pattern Detected 167 28,99 Persistent Hot Spot 135 23,44 Oscillating Hot Spot 7 1,22 Sporadic Hot Spot 16 2,78 Persistent Cold Spot 189 32,81 Sum 576 100% Persistent Hot Spot 47 8,16 Sporadic Cold Spot 2 0,35 Sporadic Hot Spot 18 3,13 Sum 576 100% Table 11: PM2.5 Afternoon Rush Hour, Emerging Hot Spot Analysis Classes 41 4.6 PM2.5 - Local Outlier 4.6.1 PM2.5 – MRH Analyzing the MRH (Figure 13, and Table 12) week days results of the local outlier analysis, we can observe that the largest category is the only Low-Low cluster (43,40%), in the western part of the study area in parishes such as Alcântara, Ajuda, Belém, Benfica, and Carnide, along with eastern regions of Parque das Nações, and Marvila. These clusters indicate areas where PM2.5 levels are low relative to their immediate temporal and spatial context. Only High-High clusters (28,65%) found in the southern parishes of the city like Beato, São Vicente, Santa Maria Maior, Penha de França, and Arroios, and in the north parishes of Lumiar, Santa Clara, and Olivais, represents areas where levels are high and similar to their surroundings. Figure 13: PM2.5 Morning Rush Hour, Local Outlier On weekends, the distribution pattern is somewhat similar. The only Low-Low cluster again forms the largest group (44,79%), suggesting that these areas maintain lower values relative to their neighbors through time. The only High-High cluster increases slightly on the weekends suggesting that certain areas continue to experience high vales compared to their neighbors, consistent with the weekday pattern. 42 The Multiple Types for week days (2,08%) and weekends (6,08%) indicates areas that do not fit into a single category and have fluctuating and less predictable values. Overall, the persistence of the Low-Low and High-High clusters suggests that there are areas in Lisbon with consistently low or higher levels in their local contexts. 4.6.2 PM2.5 – ORH ORH (Figure 14, and Table 13) week days only Low-Low clusters (43,58%) are the largest. They are present in the west part of Lisbon in the parishes of Alcântara, Ajuda, Belém, Benfica, and Carnide, and in the west in Marvila, Parque da Nações, and Olivais. These clusters indicate areas where PM2.5 levels are consistently lower than its space time neighbors. Figure 14: PM2.5 Off Rush Hour, Local Outlier Week Days Count % Weekends Count % Multiple Types 12 2,08 Multiple Types 35 6,08 Never Significant 140 24,31 Never Significant 65 11,28 Only High-High Cluster 165 28,65 Only High-High Cluster 180 36,28 Only High-Low Outlier 2 0,35 Only Low-High Outlier 8 1,39 Only Low-High Outlier 7 1,22 Only Low-Low Cluster 258 44,79 Only Low-Low Cluster 250 43,40 Sum 576 100% Sum 576 100% Table 12: PM2.5 Morning Rush Hour, Local Outlier Classes 43 Only High-High clusters (33,68%) are located in the parishes of Penha de França, São Vicente, Santa Maria, Arroios, and Santo António, as well in the north in Lumiar, Olivais, and Santa Clara. Indicating that PM2.5 values are significantly higher and similar to their surrounding areas. On the weekends, the pattern shifts slightly. Only Low-Low cluster (50,17) is bigger in this period in similar areas as the weekends, with the addition of central parishes like Alvalade, Areeiro, and parts of Lumiar. Only High-High clusters (39,73%) decrease slightly, but remain significant, especially in the downtown areas and north of Lisbon. Comparing the two periods, while Low-Low cluster remain the predominant category, there is a notable increase in these clusters during weekends. The slight decrease in High-High clusters in weekends could indicate that activities contributing to higher levels of PM2.5 are more prevalent during week days, however, the areas represented by these clusters remain largely consistent. 4.6.3 PM2.5 – ARH On week days, only Low-Low clusters (41,84%) remain the largest, in western areas such as Alcântara, Ajuda, Belém, Benfica, and Carnide, and east Alvalade, Marvila, Parque das Nações, and Olivais. These clusters identify areas where PM2.5 levels are consistently lower than their space time neighbors. Only High-High clusters (30,90%), located in downtown and central parishes. These areas are characterized by having PM2.5 levels that are significantly higher and similar to their space time surroundings. Table 13: PM2.5 Off Rush Hour, Local Outlier Classes Week Days Count % Weekends Count % Multiple Types 16 2,78 Multiple Types 14 4,69 Never Significant 98 17,01 Never Significant 77 13,37 Only High-High Cluster 194 33,68 Only High-High Cluster 177 30,73 Only Low-High Outlier 17 2,95 Only High-Low Outlier 2 0,35 Only Low-Low Cluster 251 43,58 Only Low-High Outlier 4 0,69 Sum 576 100% Only Low-Low Cluster 289 50,17 Sum 576 100% 44 During weekends, the only Low-Low cluster (41,49%) remains the largest, expanding slightly more from east to west in the central area of the city. The High-High clusters (30,03%) decreases slightly, however, northern areas remain consistent with the weekday pattern. Figure 15: PM2.5 Afternoon Rush Hour, Local Outlier Comparing the two, Low-Low clusters remain the most prominent category. The slight increase in Low-Low cluster during weekends suggests minor improvements in air quality in some central areas. High-High clusters remain consistent in the northern areas in both periods as well as downtown areas. Week Days Count % Weekends Count % Multiple Types 13 2,26 Multiple Types 36 6,25 Never Significant 135 23,44 Never Significant 113 19,62 Only High-High Cluster 178 30,90 Only High-High Cluster 173 30,03 Only High-Low Outlier 1 0,17 Only High-Low Outlier 5 0,87 Only Low-High Outlier 8 1,39 Only Low-High Outlier 10 1,74 Only Low-Low Cluster 241 41,84 Only Low-Low Cluster 239 41,49 Sum 576 100% Sum 576 100% Table 14: PM2.5 Afternoon Rush Hour, Local Outlier Classes 45 5 DISCUSSION Analyzing the concentrations of both pollutants in Lisbon during 2022, we observe that colder months, generally from October to February, tend to have higher concentration than during the hotter months. This pattern aligns with the effects of atmospheric stability, which is more present during colder months, where there are less optimal dispersion condition which can be a contributing factor for the higher concentrations measured (Zheng, 2005). Looking at the spatial distribution of PM concentrations in the city, we can see that more central and downtown areas tend to have higher concentrations than the surrounding areas, this could be the result of the effect that the urban infrastructure has on wind, which is the main component of horizontal dispersion, the presence of buildings, especially if they are denser as in these cases, it could impede the proper dispersion and result in higher concentration in those areas. Another area with constantly high values is in the northern region of the study area in the parishes of Lumiar, Olivais, and Santa Clara, which are located near the Lisbon Airport which can explain the high concentrations of PM measured in these areas. Comparing week days and weekends results of the spatiotemporal analysis of PM10 highlights certain patterns in different time periods. Week days tend to show a more consistent spatiotemporal pattern, this is can be seen by the dominance of one or two types of hot and cold spots, as well as dominance of High-High or Low-Low clusters with few outliers. This can be explained by the more linear traffic behaviors during week days. In contrast, the weekend patterns show more diverse spatial patterns, the presence of a higher number of different hot and cold spots types, as a bigger number of outliers and Multiple Types, highlight a more diverse or less predictable pattern, which can be the result of the less structured nature of weekend activities. Analyzing the spatiotemporal results for the different time periods, certain patterns can be identified. During week days MRH and ORH seem to be more consistent through time, dominated by Intensifying Hot Spots and persistent cold spots as well as fewer number of outliers compared with ARH period. ARH is represented with higher number of types of hot and cold spots, as well as, slightly more outliers. This shows 52 Lisboa Aberta, 2024. Air Quality Data. (n.d.). Retrieved February 20, 2024, from https://lisboaaberta.cm-lisboa.pt/index.php/pt/dados/conjuntos-de-dados Liu, Y., Gao, Y., Yu, N., Zhang, C., Wang, S., Ma, L., Zhao, J., & Lohmann, R. (2015). Particulate matter, gaseous and particulate polycyclic aromatic hydrocarbons (PAHs) in an urban traffic tunnel of China: Emission from on-road vehicles and gas-particle partitioning. Chemosphere, 134, 52–59. https://doi.org/10.1016/j.chemosphere.2015.03.065 Lopes, M., Russo, A., Monjardino, J., Gouveia, C., & Ferreira, F. (2019). Monitoring of ultrafine particles in the surrounding urban area of a civilian airport. Atmospheric Pollution Research, 10(5), 1454–1463. https://doi.org/10.1016/j.apr.2019.04.002 Management Association, I. R. (Ed.). (2018). Climate Change and Environmental Concerns: Breakthroughs in Research and Practice. IGI Global. https://doi.org/10.4018/978-1-5225-5487-5 Mann, H. B. (1945). Nonparametric Tests Against Trend. Econometrica, 13(3), 245. https://doi.org/10.2307/1907187 Pant, P., & Harrison, R. M. (2013). Estimation of the contribution of road traffic emissions to particulate matter concentrations from field measurements: A review. Atmospheric Environment, 77, 78–97. https://doi.org/10.1016/j.atmosenv.2013.04.028 Rovelli, S., Cattaneo, A., Borghi, F., Spinazzè, A., Campagnolo, D., Limbeck, A., & Cavallo, D. M. (2017). Mass Concentration and Size-Distribution of Atmospheric Particulate Matter in an Urban Environment. Aerosol and Air Quality Research, 17(5), 1142–1155. https://doi.org/10.4209/aaqr.2016.08.0344 Schaller, B. (2010). New York City’s congestion pricing experience and implications for road pricing acceptance in the United States. Transport Policy, 17(4), 266–273. https://doi.org/10.1016/j.tranpol.2010.01.013 53 Thorpe, A., & Harrison, R. M. (2008). Sources and properties of non-exhaust particulate matter from road traffic: A review. Science of The Total Environment, 400(1–3), 270– 282. https://doi.org/10.1016/j.scitotenv.2008.06.007 WHO (2016). Ambient (outdoor) Air Pollution Report. (n.d.). Retrieved February 20, 2024, from https://www.who.int/en/news-room/fact-sheets/detail/ambient-(outdoor)-airquality-and-health WHO (2018). Ambient (outdoor) Air Pollution Report. (n.d.). Retrieved February 20, 2024, from https://www.who.int/news-room/fact-sheets/detail/ambient-(outdoor)-airquality-and-health Zheng, M., Salmon, L. G., Schauer, J. J., Zeng, L., Kiang, C. S., Zhang, Y., & Cass, G. R. (2005). Seasonal trends in PM2.5 source contributions in Beijing, China. Atmospheric Environment, 39(22), 3967–3976. https://doi.org/10.1016/j.atmosenv.2005.03.036 Zhu, Y., Hinds, W., Kim, S.-H., & Si, S. (2002). Study of ultrafine particles near a major highway with heavy-duty diesel traffic. Atmospheric Environment, 36, 4323–4335. https://doi.org/10.1016/S1352-2310(02)00354-0 54 ANNEX Table 15A: Station ID and Location ID Location ID Location QAPM1000 01 Calçada da Ajuda QAPM1000 41 Jardim do Braço de Prata QAPM1000 02 Restelo - Rua Gonçalo Velho Cabral QAPM1000 42 Travessa de Francisco Rezende QAPM1000 03 Cais do Sodré QAPM1000 43 Avenida Almirante Gago Coutinho QAPM1000 04 Alcântara - Rua dos Lusíadas QAPM1000 44 Avenida do Santo Condestável QAPM1000 05 Avenida Vinte e Quatro de Julho QAPM1000 45 Rua Frei Carlos QAPM1000 06 Avenida Infante Santo QAPM1000 46 Entrecampos QAPM1000 07 Av Infante Dom Henrique (Chafariz Del Rei) QAPM1000 47 Avenida dos Estados Unidos da América QAPM1000 08 Baixa - Rua do Ouro QAPM1000 48 Avenida Lusíada QAPM1000 09 Praça do Comércio QAPM1000 49 Avenida de Roma QAPM1000 10 Alto da Ajuda - Rua Sá Nogueira QAPM1000 50 Chelas - Rua Dr. José Espirito Santo QAPM1000 11 Avenida de Ceuta QAPM1000 51 Avenida José Régio QAPM1000 12 Rua de São Bento QAPM1000 52 Avenida Lusíada / Qta da Granja QAPM1000 13 Rua Damasceno Monteiro QAPM1000 53 Rua Lúcio de Azevedo QAPM1000 14 Praça Martim Moniz QAPM1000 54 Avenida Marechal Gomes da Costa QAPM1000 15 Campo de Santa Clara QAPM1000 55 Avenida do Brasil QAPM1000 16 Cemitério dos Prazeres QAPM1000 56 Avenida General Norton de Matos QAPM1000 17 Jardim Botânico QAPM1000 57 Campo Grande - Museu da Cidade QAPM1000 18 Parque de Campismo de Lisboa QAPM1000 58 Jardim Professor António Franco QAPM1000 19 Monsanto - Alameda Keil do Amaral QAPM1000 59 Parque da Vinha - Estação Meteorológica QAPM1000 20 Avenida da Liberdade - Rua Manuel Jesus Coelho QAPM1000 60 Olivais Sul - Quinta Pedagógica QAPM1000 21 Rua dos Sapadores QAPM1000 61 Quinta das Conchas - Avenida Maria Helena Vieira da Silva QAPM1000 22 Campo de Ourique QAPM1000 62 Estrada do Paço do Lumiar QAPM1000 23 Avenida Almirante Reis QAPM1000 63 Estrada Militar QAPM1000 24 Rua Braamcamp QAPM1000 64 Alameda da Encarnação QAPM1000 25 Monsanto - Parque Ecológico QAPM1000 65 Avenida Doutor Alfredo Bensaúde QAPM1000 26 Parada Alto de São João QAPM1000 66 Rua Ilha dos Amores QAPM1000 27 Marquês de Pombal - Alameda Edgar Cardoso QAPM1000 67 Rua Vasco da Gama Fernandes QAPM1000 28 Beato - Avenida Infante Dom Henrique QAPM1000 68 Laboratório de Bromatologia e Águas QAPM1000 29 Avenida Fontes Pereira de Melo QAPM1000 69 Calçada de Carriche QAPM1000 30 Avenida António Augusto de Aguiar QAPM1000 70 Rua Chen He QAPM1000 31 Largo da Madre de Deus QAPM1000 71 Estrada Militar às Galinheiras 55 QAPM1000 32 Rua de Campolide QAPM1000 72 Rua Mário Botas QAPM1000 33 Largo do Leão QAPM1000 73 Rua Alferes Malheiro QAPM1000 34 Avenida da Républica QAPM1000 74 Rua da Venezuela QAPM1000 35 Praça São Francisco de Assis QAPM1000 75 Alm. P. Álvaro Proença EMQA QAPM1000 36 Estrada de Monsanto QAPM1000 76 Restauradores - Avenida da Liberdade QAPM1000 37 Praça de Espanha QAPM1000 77 Rua da Atalaia QAPM1000 38 Marvila - Rua Pedro de Azevedo QAPM1000 78 Jardim da Estrela QAPM1000 39 Estrada de Benfica QAPM1000 79 Avenida Doutor Francisco Luís Gomes / EMQA QAPM1000 40 Avenida João XXI QAPM1000 80 Rua Nau Catrineta cruz Rua Nova dos Mercadores 56 APENDIX A A1 - PM10 A1.1 Morning Rush Hour Week Days MRH WK Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Minimum 6,61 5,10 5,00 6,44 6,82 5,09 10,90 10,3 9 6,83 6,63 10,9 5 13,73 Maximum 43,7 4 41,9 5 36,5 5 47,5 8 38,4 5 71,2 1 105,0 5 41,0 0 63,5 5 48,5 0 53,8 3 129,3 4 Mean 24,3 4 26,5 6 24,2 5 16,9 5 18,8 8 18,8 7 22,07 21,9 0 22,5 2 28,3 1 27,4 8 30,96 StaDev 7,97 8,91 6,57 6,82 6,36 9,36 11,53 6,06 9,44 7,83 8,60 16,35 Median 24,4 0 29,1 8 25,2 3 17,0 0 19,0 5 18,3 4 20,81 21,4 1 20,3 3 27,8 2 27,1 7 27,52 Skewness 0,29 -0,30 -0,49 2,08 0,78 3,23 5,70 0,80 1,57 -0,15 0,70 3,97 Kurtosis 2,99 2,40 3,68 10,0 8 4,22 17,5 3 41,39 4,14 6,94 3,23 4,30 24,22 Table 16A: PM10 Morning Rush Hour Week Days Summary Statistics 57 Figure 16A: PM10 Morning Rush Hour Week Days Histogram 58 Figure 17A: PM10 Morning Rush Hour Week Days Scatter Plots Figure 18: PM10 Morning Rush Hour Week Days IDW 59 A1.2 – PM10 Morning Rush Hour Weekends MRH WE Jan Fe b Ma r Apr Ma y Ju n Jul Au g Se p Oct No v Dec Minimum 2,80 3,67 2,70 2,42 2,50 7,57 8,28 3,58 3,22 7,56 3,92 Maximum 41,6 9 48,0 6 37,3 0 37,0 0 85,3 6 38,4 4 36,2 2 39,5 6 72,0 0 39,9 4 108,1 8 Mean 19,9 1 20,3 5 17,0 9 20,4 8 17,2 8 17,3 6 22,0 3 15,7 9 22,1 7 23,5 2 24,86 StandardDeviati on 9,21 10,9 5 7,19 7,07 10,7 0 5,80 6,62 6,73 9,14 7,81 13,93 Median 18,6 5 20,8 1 17,1 0 22,1 4 16,6 8 17,0 7 22,7 2 14,4 4 21,2 3 23,8 3 22,90 Skewness 0,58 0,59 1,04 - 0,33 3,92 1,14 - 0,14 1,31 2,24 0,16 3,39 Kurtosis 2,77 2,74 4,62 3,41 25,1 3 5,15 2,74 5,67 13,5 4 2,73 21,21 Table 17: PM10 Morning Rush Hour Weekends Summary Statistics 60 Figure 19A: PM10 Morning Rush Hour Weekends Histograms 61 Figure 20A: PM10 Morning Rush Hour Weekends Scatter Plots Figure 21A: PM10 Morning Rush Hour Weekends IDW 68 A1.5 PM10 Afternoon Rush Hour Week Days ARH WK Jan Feb Ma r Apr May Jun Jul Aug Sep Oct Nov Dec Minimum 4,00 4,50 7,87 6,39 3,13 10,00 10,1 4 6,41 4,43 9,41 12,2 3 Maximum 44,0 0 48,2 6 49,9 8 44,3 0 66,2 1 103,7 3 68,0 0 57,4 4 52,0 2 50,1 6 67,4 0 Mean 24,3 2 23,5 5 18,4 5 18,2 5 17,9 8 21,88 22,3 1 21,0 1 25,6 8 25,0 7 31,3 1 StandardDeviatio n 10,1 6 10,5 6 8,15 7,67 10,4 6 12,58 9,33 11,4 7 9,55 9,28 12,5 3 Median 25,0 0 25,7 2 17,2 5 17,0 2 15,8 6 19,14 20,4 2 17,2 1 24,2 3 23,9 4 28,1 8 Skewness 0,08 0,11 1,86 1,53 2,05 4,31 2,29 1,35 0,31 0,49 0,75 Kurtosis 2,32 2,25 6,80 5,62 8,71 27,37 10,4 0 3,93 2,81 2,62 3,23 Table 20A: PM10 Afternoon Rush Hour Week Days Summary Statistics 69 Figure 28A: PM10 Afternoon Rush Hour Week Days Histograms 70 Figure 29A: PM10 Afternoon Rush Hour Week Days Scatter Plots Figure 30A: PM10 Afternoon Rush Hour Week Days IDW 71 A1.6. PM10 Afternoon Rush Hour Weekends ARH WE Jan Feb Ma r Apr May Jun Jul Aug Sep Oct Nov Dec Minimum 2,00 5,17 2,23 2,00 1,00 6,13 7,70 2,39 3,33 2,00 3,17 Maximum 47,7 6 38,3 3 32,0 8 47,0 4 48,6 1 27,8 2 26,7 0 39,1 7 38,3 0 41,7 8 164,0 0 Mean 21,4 9 17,1 3 12,3 0 15,8 9 14,9 7 14,0 1 16,9 1 10,9 4 18,4 3 19,5 1 27,31 StandardDeviatio n 10,3 3 6,02 4,65 6,50 8,14 3,89 3,97 5,74 6,07 7,31 20,70 Median 22,6 3 18,5 6 12,5 7 16,8 9 14,9 4 14,0 0 17,3 0 10,2 5 18,4 5 19,1 5 25,04 Skewness 0,07 0,37 1,02 1,31 1,91 0,95 -0,25 3,08 0,57 0,45 4,98 Kurtosis 2,28 4,08 7,41 9,24 8,47 5,26 3,28 15,9 3 4,70 4,12 32,50 Table 21A: PM10 Afternoon Rush Hour Weekends Summary Statistics 72 Figure 31A: PM10 Afternoon Rush Hour Weekends Histograms 73 Figure 32A: PM10 Afternoon Rush Hour Weekends Scatter Plots Figure 33A: PM10 Afternoon Rush Hour Weekends IDW 74 A2. PM2.5 A2.1. PM2.5 Morning Rush Hour Week Days MRH WK Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Minimum 6,46 3,03 5,68 7,08 4,00 6,17 7,45 5,07 6,69 2,13 7,98 7,38 Maximum 28,7 5 29,4 1 29,9 5 29,9 2 29,0 0 30,5 5 34,7 6 28,5 5 59,8 5 42,2 5 33,5 5 37,0 5 Mean 16,6 9 16,2 5 15,1 2 15,3 1 15,0 2 14,6 0 14,9 7 14,0 1 16,2 1 16,4 2 17,5 3 19,4 6 StandardDeviatio n 5,19 6,15 5,91 7,35 8,53 8,22 7,12 7,32 8,72 7,36 6,41 6,88 Median 14,4 5 14,1 8 12,5 0 10,9 2 10,0 0 9,82 11,5 5 11,1 9 12,5 5 13,9 4 15,1 9 16,5 8 Skewness 0,50 0,38 0,82 0,81 0,72 0,77 0,97 0,75 2,13 0,90 0,89 0,63 Kurtosis 2,18 2,14 2,62 2,03 1,80 1,89 2,67 2,12 10,1 9 3,71 2,65 2,37 Table 22A: PM2.5 Morning Rush Hour Week Days Summary Statistics 75 Figure 34: PM2.5 Morning Rush Hour Week Days Histograms 76 Figure 35A: PM2.5 Morning Rush Hour Week Days Scatter Plots Figure 36A: PM2.5 Morning Rush Hour Week Days IDW 77 A2.1.2 PM2.5 Morning Rush Hour Weekends MRH WE Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Minimum 2,20 2,08 3,50 4,35 1,08 1,05 3,70 5,22 2,17 2,22 5,25 5,14 Maximum 29,2 5 28,5 6 29,8 9 29,4 5 40,8 9 34,8 2 27,6 7 28,4 4 28,7 5 29,1 1 28,7 2 44,4 1 Mean 12,6 7 11,7 6 12,7 8 12,1 0 12,6 3 10,6 4 11,8 1 12,8 3 11,4 6 12,2 9 12,5 1 13,3 6 StandardDeviatio n 6,82 7,21 5,91 7,05 7,04 7,10 6,48 5,67 6,90 6,20 6,13 8,42 Median 9,85 8,75 11,0 0 8,30 9,39 7,45 9,61 10,8 9 9,25 10,7 1 10,0 0 9,91 Skewness 1,17 0,84 1,45 0,87 1,64 1,37 1,02 0,94 0,82 1,12 1,25 1,95 Kurtosis 3,29 2,57 4,50 2,55 5,97 4,17 3,21 3,15 2,79 3,63 3,61 6,60 Table 23: PM2.5 Morning Rush Hour Weekends Summary Statistics 84 Figure 43A: PM2.5 Off Rush Hour Weekends Histogram 85 Figure 44A: PM2.5 Off Rush Hour Weekends Scatter Plots Figure 45A: PM2.5 Off Rush Hour Weekends IDW 86 A2.5. PM2.5 Afternoon Rush Hour Week Days ARH WK Jan Feb Ma r Apr May Jun Jul Aug Sep Oct Nov Dec Minimum 3,50 3,98 4,96 4,80 3,89 6,07 4,82 4,03 1,61 4,57 8,52 Maximum 31,1 9 32,8 3 29,1 5 47,9 2 34,7 9 43,7 7 36,3 8 43,6 2 34,1 7 35,7 1 39,7 2 Mean 16,2 0 16,3 4 14,6 5 15,3 1 13,0 6 14,3 6 13,9 0 15,1 3 16,3 1 16,0 9 19,5 8 StandardDeviatio n 7,60 7,16 9,33 9,54 9,12 8,87 8,75 9,54 8,01 9,04 8,91 Median 12,5 6 13,4 8 8,35 9,88 7,51 9,47 9,09 10,1 3 13,1 0 11,1 3 15,8 2 Skewness 0,57 0,61 0,60 1,03 0,81 1,13 0,89 1,08 0,57 0,70 0,54 Kurtosis 2,04 2,29 1,44 3,16 2,00 3,17 2,23 2,86 2,04 1,90 1,84 Table 26A: PM2.5 Afternoon Rush Hour Week Days Summary Statistics 87 Figure 46A: PM2.5 Afternoon Rush Hour Week Days Histograms 88 Figure 47A: PM2.5 Afternoon Rush Hour Week Days Scatter Plots 89 A2.6. PM2.5 Afternoon Rush Hour Weekends ARH WE Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Minimum 5,70 3,28 2,60 1,17 2,56 3,30 3,33 4,15 2,04 4,30 4,11 5,53 Maximum 27,6 3 28,3 3 22,5 3 26,9 7 42,8 0 27,0 9 28,9 7 39,6 7 39,2 1 42,3 9 37,3 7 39,0 3 Mean 13,6 3 11,4 8 11,6 4 10,1 2 11,3 2 10,9 2 10,8 9 11,8 9 11,3 6 12,1 7 13,3 5 14,4 0 StandardDeviatio n 5,44 7,15 4,42 7,30 8,32 7,25 7,47 7,57 8,21 7,73 8,16 7,99 Median 12,5 7 8,17 10,3 3 6,10 8,00 7,30 6,80 8,93 7,42 9,15 10,1 1 12,1 7 Skewness 0,79 1,18 1,23 0,94 1,54 1,08 1,02 1,40 1,42 1,60 1,31 1,07 Kurtosis 2,91 3,24 4,39 2,64 5,42 2,81 2,82 4,41 4,63 5,37 3,61 3,20 Table 27A: PM2.5 Afternoon Rush Hour Weekends Summary Statistics 90 Figure 48A: PM2.5 Afternoon Rush Hour Weekends Histograms 91 Figure 49A: PM2.5 Afternoon Rush Hour Weekends Scatter Plots 92 APPENDIX – B 93 Figure 50B: 3D Visualization for PM10 Emerging Hot Spot Analysis