- 1Civil Engineering Department, Indian Institute of Technology, Delhi, New Delhi, India (cez228416@iitd.ac.in)
- 2Yardi School of Artificial Intelligence, Indian Institute of Technology Delhi, New Delhi, India 110016
Machine Learning (ML) has been widely explored for its potential in modelling air quality in numerous studies in the past. However, these approaches approximated function that maps the finite-dimensional input and output vectors. This restricts their extrapolation to unseen data and different discretization. Neural operators, a class of neural networks approximate the operator between infinite dimensional input and output functions. These models learn the underlying operator between input functions and their time-evolved state directly from data. In this work, we introduce our contribution to the field of neural operators termed Complex Neural Operator (CoNO) to learn the evolution of PM2.5 and CO concentrations over India. We trained our models using WRF-Chem simulated data over India for the years 2016-2018 and evaluated it for the year 2019. We assess our models for forecasting high pollution events, long-term forecasting (up to 72 hours) and city-level forecasts for six cities targeting two key pollutants.
How to cite: Bedi, S., Krishnan, N. M. A., and Kota, S. H.: Developing Neural Operators for Modeling PM2.5 and CO over India, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10865, https://doi.org/10.5194/egusphere-egu25-10865, 2025.