Benchmark RadarRealBench
Evaluates data-driven numerical weather forecasting models under operational conditions, using strictly out-of-distribution test data from 2025 and integrating low-latency operational analysis and large-scale in-situ observations from over 10,000 stations. It provides metrics for global forecasting and for extreme events such as heatwaves, cold surges, and tropical cyclones.
- Released
- 2026-05-24
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Existing benchmarks rely on reanalysis products that do not reflect real-time operational constraints, leading to mismatches between benchmark scores and real-world performance. This benchmark provides a more faithful and operationally relevant evaluation paradigm.
Motivation
Accurate evaluation of weather forecasting models is critical for their reliable deployment in real-world applications.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.