@ericyocam6Shipped: RAMS — AI Dataset for Methane Detection and Intensity Prediction
A research dataset supporting machine-learning-based methane detection and atmospheric tracer concentration prediction.
Developed through a multi-institutional research collaboration, the dataset uses the Regional Atmospheric Modeling System (RAMS) to simulate meteorological conditions and atmospheric tracer dispersion across a 7.6 × 7.6 km region centered on the Savannah River Site in South Carolina.
The dataset combines geospatial, temporal, and meteorological variables, including temperature, humidity, wind components, atmospheric pressure, vertical velocity, turbulent kinetic energy, and surface heat fluxes.
It supports two primary AI research tasks:
- Classification — predicting the presence or absence of a methane tracer.
- Regression — estimating atmospheric tracer concentration as a continuous measure of leak intensity.
The simulation uses a 125-meter spatial grid and covers a six-hour atmospheric modeling period. The dataset provides a foundation for evaluating machine learning and ensemble-learning approaches to environmental monitoring, industrial methane detection, and emissions analysis.
This is simulation-based research data, not direct field measurements or a production methane detection system.
The Hugging Face repository provides the dataset documentation and metadata. The underlying CSV data is hosted separately on IEEE DataPort under its standard access terms.
Dataset DOI: doi.org/10.21227/b879-d990
Comments2
Saying plainly that this is simulated data, not field readings, will spare you emails from companies wanting a ready-made leak detector. Is the next step testing it against real sensor readings from a gas utility, and do you have one lined up?
Great question, Ryan! Validating the models against real-world sensor readings would certainly be an important next step. The RAMS dataset provides a controlled simulation environment, but actual field conditions introduce additional challenges, including sensor noise, changing weather conditions, and variations in leak behavior.
I don't currently have a gas utility partner lined up for field validation, but collaborating with an industry partner would be a valuable opportunity to evaluate how well the models generalize beyond simulated data.
Ultimately, the goal is to bridge the gap between AI research and practical environmental monitoring. Appreciate the suggestion!
2 comments
Saying plainly that this is simulated data, not field readings, will spare you emails from companies wanting a ready-made leak detector. Is the next step testing it against real sensor readings from a gas utility, and do you have one lined up?
Great question, Ryan! Validating the models against real-world sensor readings would certainly be an important next step. The RAMS dataset provides a controlled simulation environment, but actual field conditions introduce additional challenges, including sensor noise, changing weather conditions, and variations in leak behavior.
I don't currently have a gas utility partner lined up for field validation, but collaborating with an industry partner would be a valuable opportunity to evaluate how well the models generalize beyond simulated data.
Ultimately, the goal is to bridge the gap between AI research and practical environmental monitoring. Appreciate the suggestion!