Paper Published in Geoscientific Model Development: Measuring Smokestack Plume Rise with Deep Learning · AUSM Lab — Augmented Urban Space Modeling Laboratory

Our joint paper with Prof. Mark Gordon’s group has been published in Geoscientific Model Development.

“Evaluation of plume rise parameterizations in GEM-MACHv2 with analysis of image data using a deep convolutional neural network”
K. M. Axelrod, M. Gordon, M. Koushafar, J. Hao, P. Makar, S. Fathi, G. Sohn. Geoscientific Model Development, 19(18), 8855–8876, 2026. doi:10.5194/gmd-19-8855-2026

How high a smokestack plume rises decides where its pollution ends up. The team used a deep learning model, developed in our lab by Dr. Mohammad Koushafar, to measure plume rise in camera images. They then used these measurements to test the plume-rise formulas in the GEM-MACHv2 air-quality model.

Steps used to measure plume rise from a camera image: the deep learning model finds the plume edges and midline, then the rise distance is measured — How plume rise is measured from a camera image (Figure 2 of the paper, Axelrod et al., 2026, CC BY 4.0).

The image data and plume masks are also openly available: Smokestack Plume Images and Plume Identification Masks (Federated Research Data Repository).

Publication

← All news