The unnecessary decline of U.S. numerical weather prediction

Overall assessment of U.S. forecast quality and the article’s framing

• Several commenters argue U.S. global models (e.g., GFS/UFS) have improved over time; the issue is lagging peers, not an actual decline.
• Some think the blog overstates the crisis: its own plots show gradual convergence with leading European models and no obvious worsening.
• Others accept that U.S. models are still behind top centers (especially the European center), and that the U.S. should aim to lead given its resources.
• One critic dislikes the nationalistic framing (“U.S. should be best at everything”), seeing it as political rather than scientific.

Politics, privatization, and NOAA’s future

• Multiple comments raise concern about efforts to shrink or break up NOAA and commercialize forecasting (e.g., Project 2025 language about downsizing and “fully commercializing” forecasts).
• There is repeated mention of private weather firms lobbying to limit public-domain data and shift value to paid services.
• Some fear political interference, loyalty tests, and replacement of career civil servants (including scientific roles) with political appointees.

Hiring, bureaucracy, and institutional dysfunction

• A detailed anecdote describes a highly qualified applicant rejected by NOAA HR for not listing “hours worked per week,” despite a director’s encouragement.
• Other federal employees confirm the process is rigid, compliance-driven, and favors insiders and veterans; hiring managers themselves often feel constrained.
• Some see this as necessary legalism; others portray HR as power-preserving and anti-meritocratic.

AI/ML vs traditional numerical weather prediction

• One side claims major centers are “stuck” in traditional NWP and not embracing AI; others directly refute this with examples of active AI forecast systems.
• Consensus in the thread: AI emulators are very cheap to run at inference time but depend on physics-based reanalysis and NWP outputs for training.
• NWP is described as irreplaceable research infrastructure, providing rich 3D physical fields that current AI models do not.

Forecast performance in practice

• Sailors and glider pilots report that high-resolution niche products (commercial or hobbyist, including ML-based) can be “mind-blowingly” accurate for micro-scale effects like island wind shadows or mountain waves.
• Others say global models like GFS are very good if interpreted with meteorological knowledge (e.g., sea-breeze, CAPE/CIN, convection not explicitly resolved).
• Some Europeans and Californians perceive worsening day-ahead rain and temperature forecasts, possibly due to fast-changing climate or loss of local observing infrastructure; this remains anecdotal and flagged as unclear.