← Back to Publications List

Unmasking Dhaka’s Air: Seasonality, Regime Shifts, and Persistence in Urban AQI (2016-2022)

Students & Supervisors

Student Authors
Saiful Islam Oni
Bachelor of Science in Computer Science & Engineering, FACULTY OF SCIENCE & TECHNOLOGY
Md. Ibtihazzaman
Supervisors
Mahfuza Khatun
Associate Professor, Faculty, FACULTY OF SCIENCE & TECHNOLOGY

Abstract

Daily air quality records for Dhaka from 2016-2022 are analysed to build an operational picture of recent conditions using AQI, PM2.5, NowCast and 24-hour averages and reported AQI categories. The daily series is aggregated to monthly values and decomposed into a month of year seasonal pattern and a smooth long term trend, allowing the strength of seasonality to be quantified while separating gradual background changes. Change-point techniques sensitive to shifts in both level and variability are then applied to the trend component to detect distinct AQI regimes over time. A clear seasonal contrast emerges: average daily AQI in winter is around 223 compared with about 100 during the monsoon, with winter often reaching the Very Unhealthy range. To characterise the persistence of poor air quality, AQI categories are mapped to an ordered state space and modelled as a discrete time Markov chain, yielding transition probabilities, steady state behaviour and expected run lengths by season. Empirically, about 48% of days fall into Unhealthy, Very Unhealthy or Hazardous categories and winter “Unhealthy or worse” spells can last for several weeks, with the longest continuous run spanning 84 days, whereas monsoon episodes typically break after only 1-2 days. High pollution behaviour is further summarised using 90th and 95th percentile AQI estimates from empirical and regression-based models, providing robust indicators of extreme events without relying on asymptotic extreme value theory. Finally, NowCast and raw based PM2.5 summaries are compared against reported AQI categories to quantify categorical agreement and misclassification rates relevant for daily health advisories. Taken together, these components provide a transparent, reproducible view of seasonal peaks, structural shifts, persistence of unhealthy air, and NowCast performance, supporting evidence based planning and clearer risk communication for urban air quality management.

Keywords

AQI PM2.5 Time series decomposition Markov chain Change point detection

Publication Details

  • Type of Publication:
  • Conference Name: International Conference on Applied Statistics and Data Science (ICASDS) 2025
  • Date of Conference: 28/12/2025 - 28/12/2025
  • Venue: University of Dhaka, Dhaka 1000, Bangladesh
  • Organizer: Institute of Statistical Research and Training (ISRT), University of Dhaka
View on Publisher