Forecasting now vs. 30 years ago

[EWR757]

I learned to forecast the weather in the 1980s and 1990s when I went to college for meteorology.  FOUS, MOS, gridded data, AVN and the ETA/NAM are some of the things I remember.  Ensembles were just starting.

 

How has forecasting changed since then?  The AFDs are certainly different now with less analysis and more synoptic descriptions.  

 

I see the models now (HRRR, NBM, ensembles, AI) are significantly more accurate than the previous generations.  If a forecaster has a different opinion than the model consensus, what tools do you use to formulate your opinion?  Can a NWS meteorologist go against model guidance?  Is there occasionally a need to go against model guidance?

 

I get the feeling in the past there was more personality and subjectivity in operational forecasting than there is today.  Is this true?

[Devidbrain]

Forecasting has certainly evolved, with high-resolution models, ensemble forecasting, and AI greatly improving accuracy. NWS meteorologists can still rely on experience and local knowledge to adjust guidance, especially in complex or rapidly changing situations. While modern forecasts are more data-driven, expert judgment remains crucial, so some subjectivity and forecaster “personality” still play a role.

[FPizz]

I haven't seen much improvement from the late 90s to now.  More models, more resolution in those models and all it created was more chaos.  With less models back 30 years ago, there was less of a guessing game to pick which model you thought was more accurate and more brain knowledge in each forecast.  There are just as many busted forecasts now as there ever was.

[Typhoon Tip]

More reliance on interpretive automation technologies. 

 

[Newman]

Well the NWS has implemented new policy/guidance that has completely stripped the human input in forecasting for days 4-7. And even days 1-3 are heavily produced based on the NBM. Take a look at the NWS Billings AFD from 3 days ago and the note at the bottom:

https://mesonet.agron.iastate.edu/wx/afos/p.php?pil=AFDBYZ&e=202604081847

So certainly the SOPs for forecasting in the NWS have completely changed. Now we're basically hitting the "send" button on whatever the NBM comes up with. And I can't emphasize enough how BAD the NBM was this past winter in many of our snowstorms. It is inherently a poor model to forecast with when you have short term model disagreement OR when you are dealing with a 90th percentile or 10th percentile type storm. The NWS Mount Holly office was still forecasting 18-24 inches of snow the starting day of the late February snowstorm out in its western CWA. And sure enough nobody saw more than 4-5" of snow. Why did this happen? Because they relied on the NBM which lags in data and model cycles AND weights certain models. Forecasting accuracy will plummet in the NWS until they figure out how to better implement the NBM and/or develop it.