Project:	Developing a Machine Learning Framework for Estimating CO₂ and Methane Emissions Using NASA Satellite Observations (JPL AO No. 16091)
Disciplines:	Computer Science, Earth Sciences
Mentor:	Sudhanshu Pandey, (JPL), Sudhanshu.Pandey@jpl.nasa.gov, Phone: (626) 379-6980
Background:	Anthropogenic carbon dioxide (CO₂) emissions from fossil fuel combustion are significantly concentrated, with nearly half originating from a handful of large point sources such as power plants. Similarly, methane (CH₄), a potent greenhouse gas, is emitted in substantial quantities from concentrated sources including landfills, oil and gas facilities, and agricultural operations. Notably, the largest power plants emit more CO₂ into the atmosphere than entire nations like Switzerland, while major methane sources play a critical role in climate forcing. Effectively monitoring and reducing emissions from these major point sources is essential for mitigating climate change. Satellite instruments offer a robust solution for global monitoring, enabling the detection and quantification of CO₂ and CH₄ emissions. For instance, NASA’s Orbiting Carbon Observatory (OCO-3) to provides high-resolution CO₂ concentration observations over emission hotspots such as cities and power plants. Additionally, instruments like the Emissions Monitoring Infrared Sounder (EMIT) enhance these efforts by globally detecting and quantifying methane emissions. Despite frequent observations of CO₂ and CH₄ plumes from large point sources in satellite data, accurately quantifying these emissions remains a complex challenge that necessitates advanced analytical tools and methodologies.
Description:	This internship project focuses on exploring the potential of machine learning approaches to quantify CO₂ and methane emissions using data from NASA’s OCO and EMIT satellite observations. The project will involve creating training and testing datasets by simulating satellite observations based on atmospheric transport model runs, which will mimic real-world conditions to provide a robust foundation for machine learning applications. A key component will be the development of a machine learning model capable of disentangling and estimating emissions from multiple CO₂ and CH₄ plumes, while accounting for variations in wind speed and direction that critically affect plume dispersion and concentration. To enhance model accuracy, additional data sources such as carbon monoxide (CO) observations from NASA’s MOPITT (Measurements of Pollution in the Troposphere) and ESA’s TROPOMI (Tropospheric Monitoring Instrument) will be integrated, helping to constrain errors related to atmospheric transport and variations in observational coverage. The model will be trained and validated using simulated CO₂, CH₄, and CO observations, aiming to achieve robust quantification of emissions even in scenarios involving overlapping plumes in complex source regions. Finally, the developed framework will be applied to real satellite data to assess its performance and refine methodologies, ultimately contributing valuable insights and tools for global emission monitoring and climate change mitigation efforts.
References:	1. Grant, D., et al : Reducing CO2 emissions by targeting the world’s hyper-polluting power plants, Environ. Res. Lett., 16, https://doi.org/10.1088/1748-9326/ac13f1, 2021. 2. Nassar, R., et al.,: Advances in quantifying power plant CO2 emissions with OCO-2, Remote Sens. Environ., 264, https://doi.org/10.1016/j.rse.2021.112579, 2021.
Student Requirements:	1. Proficiency in machine learning frameworks (e.g., TensorFlow, PyTorch or scikit-learn). 2. Familiarity with atmospheric transport models and satellite data processing is a plus.
Location / Safety:	Project building and/or room locations: . Student will need special safety training: .
Programs:	This AO can be done under the following programs:   Program    Available To        SURF@JPL    Caltech students only  Click on a program name for program info and application requirements.