Weather forecasts have become more accurate

Original Article ↗ Hacker News Discussion ↗

Weather forecasting has improved markedly over the past few decades, especially for multi‑day outlooks and severe events, thanks to better observations, physics-based models, and now emerging AI systems. Many people, however, feel short-term and hyperlocal forecasts have become less reliable, blaming factors like coarse model resolution, business incentives that favor clickbait or overconfident rain probabilities, and inconsistent quality across apps (with frequent comparisons to now-defunct Dark Sky). Commenters also highlight large regional differences, the importance of radar and local meteorologists, and the outsized impact that better forecasts could have in low-income countries where infrastructure and data are limited.

Overall perception of forecast accuracy

  • Many commenters agree long-range forecasts (3–14 days) have improved versus decades ago, citing better storm tracking and temperature trends.
  • Others report worsening or unreliable forecasts in specific locations (e.g., Boston, Bay Area, Southern California), especially for rain intensity, wind direction, or coastal/mountain microclimates.
  • Several note that 1–2 day forecasts are usually solid, but hour-by-hour or minute-level predictions feel poor.

Hyperlocal rain & “nowcasting”

  • Former Dark Sky users widely praise its minute-precise rain alerts; many feel Apple’s integration degraded accuracy.
  • Complaints about apps saying “no rain” while it’s raining (or vice versa) are frequent; explanations offered include coarse spatial grids, stale model outputs, radar gaps, and microbursts.
  • Some regions (e.g., Netherlands, parts of Europe) rely heavily on live rain radar rather than textual forecasts for short-term decisions like cycling.

Models, AI, and technical aspects

  • Discussion of physics-based Numerical Weather Prediction vs. newer AI models (GraphCast, Pangu, ECMWF’s AIFS).
  • Consensus in-thread: AI models are roughly on par or slightly better than state-of-the-art traditional models, not a dramatic leap for end users.
  • Short-range “rain in X minutes” products often use radar + simple optical flow, not deep physics.
  • Grid resolution and terrain complexity (mountains, urban heat islands) limit local accuracy.

Climate change and model reliability

  • Some suspect climate change makes forecasts worse via more volatile weather or invalidated historical data.
  • Others counter that core physics-based models remain valid; climate shifts mostly affect statistical post-processing and bias corrections, not basic forecast skill.

Human perception, probability, and media

  • Commenters highlight confirmation bias: people remember the misses, not the many quiet hits.
  • Misunderstanding of “chance of rain” leads to judging good probabilistic forecasts as “wrong.”
  • Consumer-facing providers may inflate rain probabilities or dramatize storms due to clickbait and user expectations.

Tools, strategies, and equity

  • Many recommend raw or government sources (e.g., hourly charts, radar, forecast discussions) over simplified apps.
  • Local expert meteorologists are valued for interpreting models in context.
  • The article’s point that poorer countries pay more (as GDP share) for worse forecasts is echoed as a serious equity issue.