Geri Dön

Hava kalitesi üzerindeki meteorolojik ve emisyon etkilerinin belirlenmesinde makine öğrenmesi tabanlı meteorolojik normalleştirme yönteminin uygulanması

Application of machine learning-based meteorological normalization to quantify meteorological and emissions impacts on air quality

  1. Tez No: 887355
  2. Yazar: MUHAMMED DENİZOĞLU
  3. Danışmanlar: PROF. DR. ALİ DENİZ
  4. Tez Türü: Yüksek Lisans
  5. Konular: Meteoroloji, Meteorology
  6. Anahtar Kelimeler: Belirtilmemiş.
  7. Yıl: 2024
  8. Dil: Türkçe
  9. Üniversite: İstanbul Teknik Üniversitesi
  10. Enstitü: Lisansüstü Eğitim Enstitüsü
  11. Ana Bilim Dalı: Meteoroloji Mühendisliği Ana Bilim Dalı
  12. Bilim Dalı: Atmosfer Bilimleri Bilim Dalı
  13. Sayfa Sayısı: 85

Özet

Hem antropojenik hem de doğal emisyonların bir sonucu olan hava kirliliği, insan sağlığı, çevre ve iklim üzerinde olumsuz etkileri olan küresel bir sorundur. Kirleticilerdeki mekansal ve zamansal değişiklikler meteorolojik olaylardan ve atmosferin dinamik yapısından önemli ölçüde etkilenmektedir. Bu nedenle emisyonların modellenmesinde atmosferik koşulların etkisinin çok iyi bilinmesi gerekmektedir. Bu çalışmada Türkiye'nin en gelişmiş bölgesi olan Marmara Bölgesi'ndeki 10 şehirdeki Hava Kalitesi İzleme İstasyonlarından 2013-2020 dönemine ait saatlik PM10 verilerini açıklamak için meteoroloji istasyonlarından alınan meteorolojik gözlem, sınır tabaka yüksekliği ve zamansal değişkenleri kullanılarak ağaç tabanlı makine öğrenmesi yöntemi olan random forest (RF) modelleri kurulmuştur. Marmara Bölgesi'nde meteorolojinin ve emisyonların hava kalitesindeki değişiklikler ve trendler üzerindeki etkisini belirlemek amacıyla random forest modelleri meteorolojik normalleştirme işleminde kullanıldı. Meteorolojik olarak normalleştirilmiş PM10 konsatrasyonları ile ölçüm PM10 konsantrasyonları Theil-Sen trend analizinde kullanıldı. Bu çalışmada yapılan analize göre Marmara Bölgesi'ndeki 10 şehrin tamamının yıllık ortalama değerleri Dünya Sağlık Örgütü yıllık limit değerinin çok üstündedir. Dünya Sağlık Örgütü günlük limit değerlerini aşan gün sayısı çalışma periyodu boyunca en az Çanakkale (283) ve en çok Bursa (2182) tespit edilmiştir. Model performansı determinasyon katsayısı (R2) değeri için %60 ile %73 arasında değişirken Hataların Karesinin Ortalamasının Karekökü (RMSE) değerleri için 13,85 ile 34,93 arasında değişti. Jülyen günü, sınır tabaka yüksekliği ve sıcaklık genellikle PM10 konsantrasyondaki değişimi açıklayan en önemli değişkenlerdi. Çalışma periyodu boyunca 10 şehirden 7'sinde PM10 konsantrasyonlarının uzun vadeli trendi -5,07 µgm−3 ile -0,27 µgm−3 arasındaki değerler ile azalan bir trend göstermiştir. Meteorolojik olarak normalleştirilmiş PM10 konsantrasyonlarının uzun vadeli trend değerleri 8 şehirde −0,02 µgm−3 ile −5,1 µgm−3 arasındaki değerler ile negatif bir trend göstermiştir. Yıllık ortalamalarda meteorolojinin etkisi nispeten sabit devam ederken bölgenin genelinde kirlilik emisyondan kaynaklanan konsantrasyonlar ile açıklanır. Son yıllarda ölçülen konsantrasyonlardaki düşüş azalan emisyonlar ile ilişkilidir. Bu analiz yoluyla emisyonların ve meteorolojik koşulların etkisinin sınırlarının bilinmesi bölgenin hava kalitesinin iyileştirilmesinde faydalı olacaktır. Bu çalışmanın amacı hava kirliliğini daha iyi analiz ederek hava kalitesi ve iklim değişikliği konusundaki bilgimizi arttırmaktır.

Özet (Çeviri)

Air pollution, resulting from both anthropogenic activities and natural emissions, represents a significant global issue with detrimental impacts on human health, the environment, and climate systems. The spatial and temporal variations in air pollutant concentrations are profoundly influenced by meteorological events and the dynamic nature of the atmosphere. As such, both emissions and meteorological conditions play pivotal roles in modulating air pollution levels. It is crucial to thoroughly understand the influence of atmospheric conditions when analyzing air pollution trends. Long-term observations of air pollutants from air quality monitoring stations are instrumental in assessing the efficacy of air pollution control measures and air quality policies. However, the practice of utilizing observed concentrations without considering the influence of meteorological conditions has been a subject of debate within the scientific community. Therefore, in air quality studies, particularly those related to policy analysis, it is essential to disentangle the effects of emissions from those of meteorological conditions on air pollutant concentrations. To achieve this, a range of methodologies—including traditional statistical approaches, machine learning-based techniques, and atmospheric chemistry models—have been employed in the literature. In this study, meteorological observation data—comprising wind speed, wind direction, temperature, precipitation, relative humidity, and pressure—from meteorological stations were utilized to explain hourly PM10 data for the period 2013-2020 from Air Quality Monitoring Stations in 10 cities within the Marmara Region, Turkey's most developed area. Random forest models, a tree-based machine learning method, were constructed using model output boundary layer height and temporal variables such as unix time, Julian day, days of the week, and hours of the day. Random forest is an ensemble decision tree method formed by combining multiple decision trees. It is resistant to overfitting and possesses high generalization capability. Additionally, random forest models clearly indicate the structure of each decision tree and the importance of the variables, making the internal structure comprehensible and traceable. Partial dependency graphs, which elucidate the relationships between variables, are instrumental in explaining observed trends. To optimize the accuracy and performance of the random forest model, its hyperparameters were meticulously tuned. The selected hyperparameters and their values for this study are as follows: mtry = 3, min_nod_size = 5, n_tree = 300. Random forest models were employed in the meteorological normalization process to ascertain the impact of meteorology and emissions on changes and trends in air quality in the Marmara Region. Meteorological normalization is a technique used to control or account for the meteorological impact on pollutant concentrations. In the first stage of this process, concentrations were estimated by repeatedly sampling 1000 times using models trained for each city. For the sampling process, all temporal and meteorological parameters from the original observation dataset were used as input data, excluding the trend term, unix time. This procedure resulted in 1000 forecast values for each day. Subsequently, the arithmetic mean of these 1000 predictions was calculated, referred to as the concentration level free from the effect of the current meteorology. The Theil-Sen regression technique was applied to perform long-term trend analysis of both measured and meteorologically normalized PM10 concentrations. The World Health Organization has set recommended limit values for PM10 concentrations: an annual average of 15 µg/m³ and a 24-hour average of 45 µg/m³. According to the analysis in this study, average PM10 concentrations during the study period ranged from 29.2 µg/m³ (Çanakkale) to 89.1 µg/m³ (Bursa). The annual average values for all 10 cities in the Marmara Region significantly exceed the World Health Organization (WHO) annual limit value. Regarding daily limit values, the number of days exceeding WHO limits during the study period varies between 283 (Çanakkale) and 2182 (Bursa), corresponding to 11% to 85% of the total data. The average number of exceeding days is 1241, with an average percentage of 49%. Two metrics were employed to evaluate model performance. The R² values range from 60% to 73%, with the highest R² value observed in Balıkesir and the lowest in Edirne. RMSE values range from 13.85 to 34.93, with the lowest RMSE value in Çanakkale and the highest in Bursa. This indicates that the variance in PM10 concentration in the Marmara Region can be effectively explained by meteorological measurement data, boundary layer height, and temporal variables included in the model. The correlation between predicted and observed hourly concentrations ranges from 0.75 to 0.6, with the best model performance observed in Balıkesir and Yalova. The most important variable in all provinces except Çanakkale and Istanbul was the Julian day, representing the seasonal term. Following the Julian day, boundary layer height and temperature were identified as the most significant variables explaining the variation in PM10 concentration. The subsequent variables in average variable importance were temperature, time, relative humidity, wind speed, pressure, wind direction, days of the week, and precipitation. Notably, the precipitation variable contributed minimally in all provinces except Balıkesir. PM10 concentrations significantly decreased in 7 of the 10 air quality monitoring stations in the region, with values ranging between -5.07 µg/m³ and -0.27 µg/m³. Meteorologically normalized PM10 concentrations also decreased in 8 cities, within the range of -0.02 µg/m³ to -5.1 µg/m³. The most notable increase in meteorologically normalized concentrations was observed in Çanakkale (3.61 µg/m³), while the most significant decrease was in Sakarya. The meteorological normalization technique aids in explaining the variance in PM10 concentrations. Meteorologically normalized trend estimates exhibited significantly lower uncertainty ranges compared to observations for nearly every city. Therefore, this technique assists in both accurately estimating the true trend in concentrations and reducing uncertainty in trend calculations. The explainability of random forest models, a significant advantage, was leveraged to interpret observed trends through partial dependency graphs. The effects of cold and hot seasons were evident, influenced by the seasonal component. The relationship between temperature and boundary layer height aligned with expectations, indicating that high PM10 concentrations can occur under poor dispersion conditions, high boundary layer heights, and high temperatures. While the impact of meteorology remains relatively constant on annual averages, pollution across the region is primarily driven by emissions. The observed decline in measured concentrations in recent years is associated with decreasing emissions. Understanding the limits of emissions and meteorological impacts through this analysis will be beneficial for improving air quality in the region. Furthermore, the comprehensive application of machine learning methods, particularly random forests, highlights the potential of advanced analytical techniques in environmental studies. The integration of detailed meteorological and temporal variables into predictive models underscores the importance of a multifaceted approach to air quality analysis. The findings from this study provide valuable insights into the interplay between emissions and meteorological conditions, offering a robust framework for future air quality management and policy development. This study aims to enhance our understanding of air quality dynamics and climate change by providing a detailed analysis of air pollution in the Marmara Region. By leveraging advanced machine learning techniques and comprehensive meteorological data, we can better assess the effectiveness of air quality policies and identify key factors influencing pollutant concentrations. The insights gained from this research will contribute to more informed decision-making and the development of strategies to mitigate air pollution and its impacts on public health and the environment.

Benzer Tezler

  1. Türkiye'de hava kirliliği konsantrasyon değerlerinin meteorolojikdeğişkenler ile arasındaki ilişkilerin belirlenmesi ve kirleticikonsantrasyonlarının doğası

    Determination of the relationships between air pollution concentration values and meteorological variables in Turkey and the nature of pollutant concentrations

    ENES BİRİNCİ

    Yüksek Lisans

    Türkçe

    Türkçe

    2023

    Meteorolojiİstanbul Teknik Üniversitesi

    Meteoroloji Mühendisliği Ana Bilim Dalı

    DOÇ. DR. EMRAH TUNCAY ÖZDEMİR

    PROF. DR. ALİ DENİZ

  2. Analysis of SO2 pollution in Turkey using ground observations and satellite retrievals

    Türkiye'de SO2 kirliliğinin yer ve uydu ölçümleri ile incelenmesi

    MERVE ASAR

    Yüksek Lisans

    İngilizce

    İngilizce

    2017

    Çevre Mühendisliğiİstanbul Teknik Üniversitesi

    Çevre Mühendisliği Ana Bilim Dalı

    YRD. DOÇ. DR. BURÇAK KAYNAK TEZEL

  3. Quantification of Saharan dust influences on Eastern Mediterranean air quality via atmospheric modeling

    Sahra tozunun Doğu Akdeniz hava kalitesi üzerindeki etkilerinin atmosfer modeliyle belirlenmesi

    BURCU KABATAŞ

    Doktora

    İngilizce

    İngilizce

    2016

    Meteorolojiİstanbul Teknik Üniversitesi

    İklim ve Deniz Bilimleri Ana Bilim Dalı

    PROF. DR. ALPER ÜNAL

  4. The investigation of sensitivity of copert estimated road transport emissions on air quality via WRF/CMAQ modeling system over İstanbul

    Copert4 modeliyle hesaplanan karayolları emisyonlarının duyarlılığının İstanbul hava kalitesine etkisinin WRF/CMAQ model sistemiyle belirlenmesi

    MÜGE KAFADAR

    Yüksek Lisans

    İngilizce

    İngilizce

    2015

    Meteorolojiİstanbul Teknik Üniversitesi

    İklim ve Deniz Bilimleri Ana Bilim Dalı

    PROF. DR. ALPER ÜNAL

  5. Afet bölgesinde yer alan kentlerde PM10 ve SO2 düzeylerindeki değişimlerin incelenmesi

    Analysis of changes in PM10 and SO2 levels of cities in the disaster area of Turkiye

    ALİ OZAN KURUT

    Yüksek Lisans

    Türkçe

    Türkçe

    2024

    Meteorolojiİstanbul Teknik Üniversitesi

    Meteoroloji Mühendisliği Ana Bilim Dalı

    DR. ÖĞR. ÜYESİ CEYHAN KAHYA