ericyoc/rams-methane-tracer-dataset

RAMS-Simulated Meteorological and Atmospheric Tracer Dataset for Ensemble Learning-Based Methane Detection and Intensity Prediction

This repository is a dataset card only. The data itself is hosted on IEEE DataPort.

DOI: 10.21227/b879-d990 Access: IEEE DataPort Standard Dataset — an IEEE DataPort subscription is required to download the files. IEEE Society members have access through their membership. No data is redistributed here.

Description

RAMS (Regional Atmospheric Modeling System)-simulated meteorological and atmospheric tracer concentration data, generated for ensemble learning-based methane (CH₄) detection and intensity prediction.

The simulation covers a 125 m grid over a 7.6 × 7.6 km domain centered on the Savannah River Site (SRS), Aiken, South Carolina, USA, running for six hours from 10 February 2022, 1200 UTC.

Contents

Spatiotemporal grid indices, geographic coordinates, and the following meteorological variables:

Group	Variables
Thermodynamic	temperature, relative humidity, pressure, water vapor
Dynamic	vertical velocity, turbulent kinetic energy, u_west_to_east_wind, v_south_to_north_wind
Surface flux	sensible heat flux, latent heat flux, precipitation rate
Spatial / temporal	i_value, j_value, latitude, longitude, Time (decimal UTC hours)
Target	tracer_concentration — atmospheric tracer representing fugitive methane

Derived predictors documented with the dataset include wind speed, directional sine and cosine embeddings, and diurnal periodic features.

Format: CSV, 13.26 MB.

Intended Tasks

Classification — methane detection. Binarize the target:

df['methane_present'] = (df['tracer_concentration'] > 0).astype(int)

Regression — intensity prediction. Use tracer_concentration directly as a continuous target.

feature_cols = [
    'temperature', 'relative_humidity', 'vertical_velocity',
    'pressure', 'water_vapor', 'turbulent_kinetic_energy',
    'precipitation_rate', 'sensible_heat_flux', 'latent_heat_flux',
    'i_value', 'j_value', 'latitude', 'longitude',
]

Modeling Notes

• Target leakage. Always exclude tracer_concentration from the feature set.
• Class imbalance. The classification task contains structural zeros. Use stratified splitting and metrics appropriate to imbalance (F1, AUC-ROC).
• Spatial autocorrelation. Observations are indexed by i_value, j_value, latitude and longitude; neighbouring cells are not independent.
• Temporal structure. Time spans a 6-hour window in decimal UTC hours; temporal ordering matters for time-series approaches.

Preprocessing applied by the authors: consistent variable naming and typing, duplicate record checks, and missing-value checks.

Citation

@data{b879-d990-26,
  doi       = {10.21227/b879-d990},
  url       = {https://dx.doi.org/10.21227/b879-d990},
  author    = {Anjana Rajendra Prasad and Nathaniel Ketema and Eric Yocam and
               Varghese Vaidyan and Gurcan Comert and David Werth and
               Robert Buckley and Milinda Rambel Stone and Baby Vennela Kothakonda},
  publisher = {IEEE Dataport},
  title     = {RAMS-Simulated Meteorological and Atmospheric Tracer Dataset for
               Ensemble Learning-Based Methane Detection and Intensity Prediction},
  year      = {2026}
}

Authors

Anjana Rajendra Prasad (Dakota State University) · Nathaniel Ketema (Dakota State University) · Eric Yocam (Dakota State University) · Varghese Vaidyan (Dakota State University) · Gurcan Comert (North Carolina A&T State University) · David Werth (Savannah River National Laboratory) · Robert Buckley (Savannah River National Laboratory) · Milinda Rambel Stone (Dakota State University) · Baby Vennela Kothakonda (Dakota State University)

Submitted 17 March 2026.