Open Conference Systems, DDAYS LAC 2024 Main Conference

Font Size: 
Enhancing Real-Time Air Quality Modeling: A Machine Learning Approach to Downscale WRF-Chem
Lucas Luciano Berná, Rafael Pedro Fernandez, Emmanuel Nicolas Millan, Enrique Puliafito

Building: Cero Infinito
Room: Posters hall
Date: 2024-12-10 04:30 PM – 06:30 PM
Last modified: 2024-11-19

Abstract


The Weather Research and Forecasting Model with coupled Chemistry (WRF-Chem) is widely used for air quality studies due to its ability to resolve atmospheric dynamics and chemical processes simultaneously. While WRF-Chem provides highly detailed simulations, its computational cost is significantly higher—between 100 and 1,000 times more expensive than regional models without chemistry. This presents a challenge for its application in real-time, high-resolution operational platforms. In this work, we address this limitation by introducing a machine learning approach to enhance the operational efficiency of WRF-Chem. Our method uses high-resolution WRF-Chem simulations, computed without time constraints and using the GEAA emissions inventory for Argentina, to train a neural network. This network is then applied to improve the resolution of lower-cost, coarse simulations, allowing for real-time forecasting with significantly reduced computational demand. The focus of this study is on evaluating the final model's ability to downscale simulation resolution while minimizing operational times. By comparing WRF-Chem simulations with and without this machine learning enhancement, we demonstrate the substantial gains in computational efficiency, offering a practical and cost-effective solution for real-time air quality forecasting.