Climate Benchmark · nasa-impact
WxC-Bench
A standardised benchmark for evaluating AI models across six atmospheric and earth science downstream tasks.
What WxC-Bench is
WxC-Bench is a multi-task atmospheric and earth science benchmark developed by NASA IMPACT and collaborators. It provides standardized datasets across six meteorological challenges: nonlocal parameterisation of gravity wave momentum flux, aviation turbulence prediction, weather analog search, natural language weather forecast generation, long-term precipitation forecasting, and hurricane track and intensity prediction.
The resource addresses the need for a unified evaluation standard for artificial intelligence models designed for complex meteorological phenomena. By gathering disparate operational, reanalysis, and observational sources into task-specific datasets, it allows researchers to systematically measure and compare model performance across physical forecasting and atmospheric analysis domains.
How it was made
The benchmark assembles data from multiple atmospheric and satellite records, including ERA5 reanalysis, High-Resolution Rapid Refresh (HRRR) reanalysis, NOAA Storm Prediction Center reports, HURDAT2, PERSIANN CDR, IMERG final daily products, and satellite brightness temperature CDRs (PATMOS-x, GridSat-B1, and SSMI(S)). Baseline forecast comparisons for extended precipitation forecasting incorporate data from the ECMWF and UK Met Office S2S databases.
Data curation involves spatial and temporal alignment, quality control, and variable normalisation. Individual tasks have specific processing; for example, the gravity wave momentum flux dataset derives from ERA5 winds and temperatures coarsely grained on a 0.3° grid with the top 15 pressure levels above 1 hPa excluded, concatenating dynamic variables across the vertical dimension to target vertical momentum flux components.
Key facts
- Curator
- NASA IMPACT and contributing researchers
- Number of tasks
- 6
- Underlying data sources
- ERA5, HRRR, HURDAT2, PERSIANN CDR, IMERG, NOAA SPC reports, PATMOS-x, GridSat-B1, SSMI(S)
- Forecast baselines
- ECMWF and UK Met Office S2S database
- DOI
- 10.57967/hf/7711
What you can use it for
- Gravity Wave Parameterisation
- Evaluate models predicting vertical components of gravity wave momentum flux from vertically concatenated zonal wind, meridional wind, and potential temperature profiles.
- Aviation Turbulence Detection
- Test models designed to identify and predict hazardous atmospheric turbulence conditions affecting aviation safety.
- Extended Precipitation Forecasting
- Assess global daily rainfall predictions up to 28 days ahead using satellite observations and historical precipitation records.
- Hurricane Track and Intensity Analysis
- Benchmark machine learning algorithms predicting tropical cyclone trajectories and intensity shifts using HURDAT2 records.
- Automated Weather Forecast Text Generation
- Train and evaluate models generating human-readable forecast discussions from paired HRRR reanalysis data and NOAA Storm Prediction Center reports.
- Weather Analog Identification
- Assess pattern-matching methods that detect historically similar weather states to aid forecasting.
Explore the data
No queryable preview
This dataset is not available in a queryable form — usually an unusual file format or a repository too large to convert. The original source may still be readable.Strengths and limits
Good at
- Provides standardized evaluation data across six distinct meteorological and atmospheric science tasks.
- Integrates curated observational and reanalysis datasets from authoritative earth science sources including ERA5, IMERG, and NOAA records.
- Includes established baseline forecasts from ECMWF and UK Met Office S2S for long-term precipitation assessment.
Watch out for
- Gravity wave data excludes the top 15 ERA5 pressure levels above 1 hPa.
- The natural language forecast generation task is constrained to January 2017 reports and reanalysis.
- Long-term precipitation data relies on PERSIANN CDR which terminates after June 2020.
Questions
Can I look inside WxC-Bench without downloading it?
Not here: it is not available in a queryable form. Its description and statistics are still shown.
What licence is WxC-Bench under?
Its authors declare mit. Read the terms on the original page before reusing the data.
What tasks are included in WxC-Bench?
It includes six tasks: nonlocal parameterisation of gravity wave momentum flux, aviation turbulence prediction, weather analog search, natural language forecast generation, long-term precipitation forecasting, and hurricane track and intensity prediction.
What are the inputs and outputs for the gravity wave task?
Inputs are three atmospheric variables—zonal wind, meridional wind, and potential temperature—concatenated along the vertical dimension. The outputs are the zonal and meridional components of vertical momentum flux resulting from gravity waves.
How far in advance does the precipitation forecasting task extend?
The long-term precipitation forecasting task targets predictions up to 28 days in advance using daily global rainfall accumulation records and satellite observations.
What sources are used for the hurricane and language forecasting tasks?
Hurricane track and intensity prediction uses the HURDAT2 dataset, while natural language weather forecast generation pairs HRRR reanalysis data with NOAA Storm Prediction Center daily reports.
How to cite it
Rajat Shinde, Kumar Ankur, Christopher E. Phillips, Aman Gupta, Simon Pfreundschuh, Sujit Roy, Sheyenne Kirkland, Vishal Gaur, Venkatesh Kolluru, Amy Lin, Prajun Trital, Aditi Sheshadri, Udaysankar Nair, Manil Maskey, Rahul Ramachandran. WxC-Bench (Revision d0910b2), 2026. DOI: 10.57967/hf/7711
Written from WxC-Bench's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.