TROPOMI/S5P sensitivity limits with respect to detection of NO2 plumes from seagoing ships

This is a Preprint and has not been peer reviewed. The published version of this Preprint is available: This is version 2 of this Preprint.

Add a Comment

You must log in to post a comment.


There are no comments or no comments have been made public for this article.


Download Preprint


Solomiia Kurchaba, Artur Sokolovsky, Jasper van Vliet, Fons J. Verbeek, Cor J. Veenman


The marine shipping industry is among the strong emitters of nitrogen oxides (NOx) -- a substance harmful to ecology and human health. Monitoring of emissions from shipping is a significant societal task. Currently, the only technical possibility to observe NO2 emission from seagoing ships on a global scale is using TROPOMI data. A range of studies reported that NO2 plumes from some individual ships can be visually distinguished on selected TROPOMI images. However, all these studies applied subjectively established pre-determined thresholds to the minimum speed and length of the ship -- variables that to a large extent define the emission potential of a ship. In this study, we investigate the sensitivity limits for ship plume detection as a function of their speed and length using TROPOMI data. For this, we train a classification model to distinguish TROPOMI image patches with a ship, from the image patches, where there were no ships. This way, we exploit ground truth ship location data to potentially exceed human visual distinguishability. To test for regional differences, we study four regions: the Mediterranean Sea, Biscay Bay, Arabian Sea, and Bengal Bay. For the Mediterranean and the Arabian Sea, we estimate the sensitivity limit to lie around a minimum speed of 10 knots and a minimum length of 150 meters. For the Biscay Bay - around 8 knots and 100 meters. We further show that when focusing the analysis on the biggest emitters (junctions of several ships in the area), the detectability can be improved up to above 0.8 ROC-AUC. Finally, we show that increasing the size of the dataset, beyond the dataset used in this study, yields further improvements in the detectability of smaller/slower ships. The rate of improvement in both experiments is dependent on the region studied. This paper is the first comprehensive study on the real-world sensitivity of the TROPOMI instrument to distinguish the NO2 emission produced by seagoing ships.



Artificial Intelligence and Robotics, Atmospheric Sciences, Environmental Monitoring


TROPOMI sensitivity limits, machine learning, Emissions, seagoing ships, NO2, TROPOMI, sensitivity limits


Published: 2023-09-20 13:45

Last Updated: 2024-02-15 19:53

Older Versions

CC BY Attribution 4.0 International

Additional Metadata

Conflict of interest statement:

Data Availability (Reason not available):
Dataset used in the study will be available upon publication of the article