Project:	Deep Learning for Operational Greenhouse Gas Plume Monitoring with the EMIT Imaging Spectrometer (JPL AO No. 16713)
Disciplines:	Computer Science, Remote Sensing, Imaging Spectroscopy
Mentor:	Jake Lee, (JPL), jake.h.lee@jpl.nasa.gov, Phone: (818) 354-2578
Mentor URL:	https://ml.jpl.nasa.gov/members/jake-lee.html  (opens in new window)
Background:	Anthropogenic emissions of methane and carbon dioxide account for the vast majority of the global warming contribution. A small percentage of these emission sources dominate total emissions; therefore, locating these sources and quantifying their emissions improves our understanding of overall anthropogenic emissions and unlocks efficient and impactful mitigation opportunities. For nearly a decade, we have developed and validated GHG plume detection systems driven by spaceborne (EMIT, Carbon Mapper) and airborne (AVIRIS, GAO) imaging spectrometers. These Earth-observing sensors capture visible to shortwave infrared (VISWIR) wavelengths, which allows us to retrieve CH4 and CO2 concentrations in the atmosphere. Our prior works have demonstrated robust detection of methane plumes using CNNs and UNets; our current focus is on deploying such models operationally on the EMIT mission, and on investigating newer deep learning methods to improve detection accuracy and enable additional capabilities. Developing a model that generalizes globally demands thorough and rigorous analysis of capabilities and limitations on spatiotemporally diverse observations under a wide range of imaging conditions.
Description:	The goal of this project is to deploy a robust GHG plume detection system for the EMIT mission and other imaging spectrometers. There are several related parallel concepts that contribute towards this goal. Students may propose to contribute to one or more of the following ongoing and future research concepts, or other concepts as they arise: - Train and evaluate plume detection models on multiple gas species (e.g. CH4 and CO2) or instruments (e.g. EMIT and AVIRIS), instead of needing to train a separate model for each gas and instrument. - EMIT makes repeat observations of the same region, and some emissions are persistent, while others are intermittent. Investigate whether including past observations may improve detection performance. - The input to these models are maps of gas concentration retrieved by applying a CMF to radiance spectra to distinguish methane absorption features from noise and surface materials. Prior work has shown that deep learning models could retrieve gas concentrations directly from radiance spectra. Train a plume detection model directly on the spectra with SpecTf, a transformer model. - Some infrastructure (oil and gas, landfills, agricultural, etc.) are visibly co-located with GHG plumes, whereas others are not (buried pipelines, etc.) Investigate whether including infrastructure information may improve detection performance. For all concepts, students will work closely with an interdisciplinary team of subject matter experts in machine learning, atmospheric science, and imaging spectroscopy to guide decision-making.
References:	PBS NewsHour Story: https://www.pbs.org/newshour/show/nasa-scientists-track-climate-changing-methane-leaks-from-the-air EMIT VISIONS Open Data Portal: https://earth.jpl.nasa.gov/emit/data/data-portal/Greenhouse-Gases/ Relevant Publications: - Bue et al. preprint: https://arxiv.org/abs/2505.21806 - Ruzicka et al. preprint: https://arxiv.org/abs/2511.07719 - Lee and Keely, Statistical Learning in Atmospheric Chemistry: https://www.youtube.com/watch?v=-a4JpeZK0bU - Lee et al., 2025 PNAS: https://doi.org/10.1073/pnas.2502903122 - Duren et al. 2019 Nature: https://www.nature.com/articles/s41586-019-1720-3 - Thompson et al. 2015 AMT: https://amt.copernicus.org/articles/8/4383/2015/ Publications by previous interns: - Mancoridis et al., 2025 IEEE TGRS https://doi.org/10.1109/TGRS.2025.3608601 - Hu et al., 2025 AGU Fall "Sensitivity and Uncertainty Aware Deep Learning for Improved EMIT Methane Plume Detection" - Wei et al. 2025 AGU Fall "Segment Anything Model for EMIT Methane Plume Delineation and Mask Refinement" - Satish et al. 2023 AGU Fall "Improving Deep Learning Methods for Robust Methane Plume Detection using Alternative Input Representations" - Mancoridis et al. 2023 AGU Fall "Leveraging Airborne Data to Enable Spaceborne Methane Plume Detection via Model and Data Driven Approaches" - Rao et al. 2021 AGU Fall "Improving Imaging Spectrometer Methane Plume Detection with Large Eddy Simulations"
Student Requirements:	Required: Coursework in machine learning/deep learning/computer vision, experience with Pytorch and scientific python (numpy, pandas, scipy, scikit-learn, etc.) Optional: Familiarity with remote sensing raster data and python-based geospatial analysis libraries (e.g. GDAL, rasterio, geopandas). Familiarity with developing on remote systems, navigating bash, and high performance computing w/ GPUs.
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    both Caltech and non-Caltech students  Click on a program name for program info and application requirements.