Session Presentations

Alqamah Sayeed, Yunsoo Choi, Ebrahim Eslami, Jia Jung, Yannic Lops, and Ahmed Khan Salman

Issues regarding air quality and related health concerns have prompted this study, which develops an accurate and computationally fast, efficient hybrid modeling system that combines numerical modeling and machine learning for forecasting concentrations of surface ozone. Currently available numerical modeling systems for air quality predictions (e.g., CMAQ, NCEP EMP) can forecast 24 to 48 hours in advance with sufficient accuracy. In this study, we develop a modeling system based on a convolutional neural network (CNN) model that is not only fast but covers a temporal period of two weeks with a resolution as small as a single hour for 255 stations. The CNN model uses forecasted meteorology from the Weather Research and Forecasting model (processed by the Meteorology-Chemistry Interface Processor), forecasted air quality from the Community Multi-scale Air Quality Model (CMAQ), and previous 24-hour concentrations of various measurable air quality parameters as inputs and predicts the following 14 day hourly surface ozone concentrations. The model achieves an average accuracy of 0.91 in terms of the index of agreement for the first day and 0.78 for the fourteenth day while the average index of agreement for a one day ahead prediction from the CMAQ is 0.77. Through this study, we intend to amalgamate the best features of numerical modeling (i.e., fine spatial resolution) and a deep-neural network (i.e., computation speed and accuracy) to achieve more accurate spatio-temporal predictions of hourly ozone concentrations. Although the primary purpose of this study is the prediction of hourly ozone concentrations, the system can be extended to various other pollutants.

Khanh Do, Arash Kashfi Yeganeh, Cesunica E. Ivey

Chemical and Environmental Engineering, University of California, Riverside

Center for Environmental Research and Technology, University of California, Riverside

California's South Coast Air Basin (SoCAB) is well-known for poor air quality due to its unique terrain and high levels of anthropogenic emissions. In this study, we use machine learning (ML) to recognize the natural pattern of ambient air pollutants in SoCAB and to explore the link between precursor emissions, meteorology and PM_2.5/ozone. We investigated the historical changes of PM_2.5 and ozone using 25 years of air pollutant, emissions, and meteorology data. We tested the random forest regression (RFR) algorithm under multiple configurations to tune the prediction and provide the best air quality predictions. We first trained the RFR model with hourly meteorology and air pollutant data from 1994 to 2018. Meteorological data were retrieved from Ontario and Los Angeles International Airport monitoring stations, and air quality data were retrieved from the Fontana (CA) air monitoring station. The RFR training features were NO, NO₂, surface temperature, dew point temperature, visibility, surface pressure, relative humidity, wind speed, and wind direction. The RFR model was trained in five-year increments from 1994 to 2018. The R² ranged from 0.6 - 0.7 for historical hourly predictions. The model also enabled predictions of 2023 PM_2.5 and ozone using input data from a 2023 CMAQ simulation. The freedom of choosing input features enabled the investigation of PM_2.5 and ozone sensitivity to fluctuations in key variables, such as temperature and NO_X. These promising results indicate that ML can accelerate air quality research by augmenting traditional air quality modeling, reducing simulation time, and exploiting large datasets for historical simulations and future air quality predictions.

Samuel H. Jones, Kun Zhang, Nicholas Muller, Peter Adams

Machine learning techniques have been employed to discover insights, solve and accurately predict many highly complex problems in the past two decades. This work applied these advancements to atmospheric pollution modeling. This paper represents a step toward applying machine learning to mimic chemical transport models (CTM). The ultimate goal is to develop a machine learning-based atmospheric chemistry model that is both accurate and computationally efficient. This paper took one small step toward that goal by developing a machine learning algorithm that: learns causes and effects of three chemical species from the area emission inputs and concentration outputs of a CTM, lays a groundwork for learning other atmospheric chemical interactions, and reduces computational costs for simulating atmospheric pollution to make these simulations more accessible to policymakers on a standard personal computer with a modest GPU. This work does not provide an adequate substitute for a CTM for our target audience but represents a step in that direction.

Alican Mertan^a, Alper Unal^b

^a Istanbul Technical University, Computer Engineering Department, Istanbul, Turkey ^bIstanbul Technical University, Eurasia Institute of Earth Sciences, Department of Climate and Marine Science, Istanbul, Turkey

Machine learning is the study of algorithms that allows machines to learn through experience. With the recent advances in computing power and the availability of the large-scale data, deep learning methods (a sub eld of machine learning) started to achieve successful results on a diverse set of problems and gain lots of popularity. One of the many areas where deep learning achieved great success is forecasting. In this work, we investigate the usage of fully connected feed forward neural networks, also known as multilayer perceptron (MLP), recurrent neural networks (RNN) and its variants. We formulate the air quality forecasting as a regression problem where the previous hourly PM10 measurements are used to estimate the future PM10 values. We compare the performance of different models and settings on our dataset, which consists of hourly PM10 measurements of Adana from 2009 to 2019. This paper presents comparison of results using different Deep Learning Techniques.