Weather Forecasting Still Feels Reactive, not Proactive

[RichmondTarHeel]

I'm bored at work, having already mastered Machine Learning 101 and waiting for the next lesson to come out.  So, I got to thinking about weather forecasting, and how machine learning can hopefully one day greatly improve it (if it hasn't already).

Now, I'm definitely not a pro forecaster or meteorologist.  In fact, I know very, very little about how anything works.  Instead, all of my knowledge comes from reading on various forums with you experts (some real, some imagined).  However, I've noticed something:  Even though weather models in the last decade have made vast improvements, forecasts are still entirely reactionary based on the latest model.  Forecasts, especially 3+ days out, swing back and forth.  Weather models have done a great job of predicting some snowstorms (like Jan 2016) several days out, but too often they go from showing a HECS to rain, back to snow, then to sunny skies - usually every 6-12 hours.  Forecasters then react to what the model has just shown, rather than predict what the next model run will look like.

I know there are probably too many unknowns to ever truly predict the weather with great accuracy, but I'm curious to know, with the advances in Machine Learning and Data Analytics, what's in store for forecasting and weather modeling in the future?  I'm sure that agencies are already using this technology to improve their models and forecasting, but how far away are we from seeing an impact?  Are they hiring talented computer and data scientist to help?  What's weather forecasting going to look like in 5 years?

[Bob Chill]

Modeling physics is pretty good at representing the atmosphere. The main hurdle is data at initialization. If models had a perfect representation of the global atmosphere then skill would go way up in the med range. IMHO- the "breakthrough" will come when satellites are capable of accurately sampling the globe. We're a long way away from that. There are many data sparse holes in the fabric. I'll prob be long gone before weather is modeled accurately in the d7 range. 

[isohume]

Weather forecasting is all about individual certainty gained from numerical guidance, statistics, climo, theory, and most importantly operational experience. It can be highly subjective. A forecast should be based on and relayed to others with confidence levels in mind either implictly or explicitly. The idea that model guidance needs to get better before forecasting is improved is missing the point. There will always be inherent pyhsical limitations to model output. Forecasting is improved through research, case studies, increased pattern recognition, training, etc. Folks expecting the models to nail a forecast are in for a wild ride, which is whats seen here with the constant frustration of chasing specific model output.