Original title: WIND-Bench: A Benchmark Dataset for In-Situ Near-Surface Wind Speed Observations Across the Conterminous United States
Kyla Bazlen, Grant Buster, Brandon Benton, Lauren North +4 · September 2026
Weather forecasts often miss strong winds near the ground in rough terrain like hills and mountains. This is a problem because accurate wind forecasts protect public safety and guide big operational decisions. Machine learning can beat traditional weather prediction models, but researchers lacked a shared dataset to test these models fairly across the Conterminous United States. To fix this, scientists created a benchmark dataset called WIND-Bench. It pulls together ground-level measurements from many sensor networks inside the Meteorological Assimilation Data Ingest System. To clean the data, the system checks readings against forecasts from NOAA's High-Resolution Rapid Refresh model. This check tells the difference between broken sensors and real high winds, providing a reliable standard to evaluate wind forecasts.
Imagine you're trying to tell if a friend's phone call is breaking up because of bad reception or because their phone is broken. You can't easily tell the difference, so you might blame the wrong thing. In this paper, the researchers faced the same problem with wind sensors: is a weird reading a real gust or a faulty sensor? They used a trusted weather forecast to check, just like you might call from a different phone to test the line.
Accurate wind forecasts are critical for aviation, wildfire management, and wind energy. But forecasts often miss dangerous gusts in hilly or mountainous areas. This dataset gives researchers a standard way to test and improve models, so those forecasts can get better.
The abstract doesn't explicitly list limitations, but it indicates the dataset covers only the conterminous United States and uses the MADIS network and HRRR model for validation. It may not generalize to other regions or forecast systems without adaptation.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
WIND-Bench: A Benchmark Dataset for In-Situ Near-Surface Wind Speed Observations Across the Conterminous United States — Kyla Bazlen, Grant Buster, Brandon Benton et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.