will meterologists become extint?

[rdale]

Mark - I'm not following your claim (especially in the second paragraph.) Why would you say that a computer can't watch every storm but a human can? Experience shows the opposite - in big events, humans miss some signs because of the info overload. Computers don't suffer from that... And as Jeff noted - computers already do outforecast humans in the short term model department.

 

[Mark Ellinwood]

Mark - I'm not following your claim (especially in the second paragraph.) Why would you say that a computer can't watch every storm but a human can? Experience shows the opposite - in big events, humans miss some signs because of the info overload. Computers don't suffer from that... And as Jeff noted - computers already do outforecast humans in the short term model department.

Picture if you will...

A computer model pops up a storm 10 miles northwest of Kansas City at 6pm and slides it eastward, staying just north of the city. It found a pocket of higher instability in that area based on a site whose temperature and dew point was running too high, allowing the storm to form there. 6pm rolls around, and the storm forms, but it's actually 10 miles further south when it first pops up on radar. A meteorologist puts out a warning for downtown KC, whereas it would have been north of the city if you were just using the straight model output.

A computer cannot nowcast. It must make complex calculations and then spit out a graphic that people can read and understand. A human can see the storm pop up and can make a nowcast accordingly, as humans do not need to make complex calculations and can follow the storm on the fly.

Computers do NOT already out-forecast humans in the short term department. In fact, it's still really easy to beat the models in instances where model resolution isn't good enough to properly handle 1) a storm between grid points, or 2) handling variables within areas that don't have a good enough topographic resolution. Where the heck is the data to support the models being better than human forecasters?

Companies pay private forecasters thousands of dollars per year for short term forecasts and nowcasts. Why on Earth would they be paying for that if there was a model that could do a better job?

 

[Jeff Duda]

Companies pay private forecasters thousands of dollars per year for short term forecasts and nowcasts. Why on Earth would they be paying for that if there was a model that could do a better job?

Because they are not aware that NWP models with the capability of forecasting individual thunderstorms exist. OR (and this makes a lot of sense from the standpoint of a private sector business owner in the business of forecasting the weather) companies that know about these models haven't told their clients. Why tell your client there's a free way to get almost the same information they pay you for?

Your example is a good one, but it relies on the assumption that a storm has already formed (i.e., advection is generally the only governing force the human uses for the forecast). What about situations requiring accurate forecasts of convective initiation? A human will most certainly be unable to beat an NWP model in that situation.

 

[Mark Ellinwood]

Because they are not aware that NWP models with the capability of forecasting individual thunderstorms exist. OR (and this makes a lot of sense from the standpoint of a private sector business owner in the business of forecasting the weather) companies that know about these models haven't told their clients. Why tell your client there's a free way to get almost the same information they pay you for?

Clients are always looking for a way to cut costs. Forecast models are shared endlessly on social media. If there was something for them to catch onto, I'm confident that they would have. There's no massive conspiracy of private forecasters withholding verification data in order to cover their own asses. I take GREAT pride in the fact that I (like most other mets) can beat NWP at any given lead time. I'm an honest, blunt person... if NWP consistently beat my forecasts, I'd quit forecasting.

Your example is a good one, but it relies on the assumption that a storm has already formed (i.e., advection is generally the only governing force the human uses for the forecast). What about situations requiring accurate forecasts of convective initiation? A human will most certainly be unable to beat an NWP model in that situation.

That's not the point I was trying to make. The point was that, no matter how much the NWP improves, there will always be a way for forecasters to find ways to beat the models. It might not be for a forecast 2-3 hours from now, but in the near term where the minutes count and in the longer range where skill declines, there's always added skill to be had through human input.

 

[rdale]

it's actually 10 miles further south when it first pops up on radar. A meteorologist puts out a warning for downtown KC, whereas it would have been north of the city if you were just using the straight model output.

I also picture the computer recognizing the measurement error, and correcting for it when the warning is issued. How do you think the meteorologist knew there was an error? His training and experience. Not some "feeling" in his gut. Well the computer can be trained to recognize errors too, and also compensate for said errors.

A computer cannot nowcast.

As has been noted already - computers do. They are still early in the development cycle, but they can take a storm that has formed (and even one that has not) and "nowcast" from that. I'd suggest looking up "Warn On Forecast" for more info on what exists now and where it's going. http://journals.ametsoc.org/doi/abs/10.1175/2009BAMS2795.1

It must make complex calculations and then spit out a graphic that people can read and understand.

No - it doesn't at all. It can spit out an ALL CAPITAL TORNADO WARNING if it wants, or it can notify cell phones in the area of the storm that a tornado is likely to form, and post a note on Twitter. Long before the human met can do all those.

as humans do not need to make complex calculations and can follow the storm on the fly.

And that's a detriment - not a feature The computer can analyze EVERY storm on the radar scope, and storms that aren't even on the screen yet, and see which ones are moving into better/worse environments. A human can do a handful, but if you have 12 supercells out there - the PC will win.

Companies pay private forecasters thousands of dollars per year for short term forecasts and nowcasts. Why on Earth would they be paying for that if there was a model that could do a better job?

Actually those private forecasters are using models that people outside of their environment don't have access to. It's cheaper to pay AW $100K per year than it is to develop a $100M modeling system.

I appreciate your concern for the plight of the human forecaster but I'm just not sure you are up to speed on where things are now (and what that means for the future.)

 

[Mark Ellinwood]

I also picture the computer recognizing the measurement error, and correcting for it when the warning is issued. How do you think the meteorologist knew there was an error? His training and experience. Not some "feeling" in his gut. Well the computer can be trained to recognize errors too, and also compensate for said errors.

As has been noted already - computers do. They are still early in the development cycle, but they can take a storm that has formed (and even one that has not) and "nowcast" from that. I'd suggest looking up "Warn On Forecast" for more info on what exists now and where it's going. http://journals.ametsoc.org/doi/abs/10.1175/2009BAMS2795.1

No - it doesn't at all. It can spit out an ALL CAPITAL TORNADO WARNING if it wants, or it can notify cell phones in the area of the storm that a tornado is likely to form, and post a note on Twitter. Long before the human met can do all those.

And that's a detriment - not a feature The computer can analyze EVERY storm on the radar scope, and storms that aren't even on the screen yet, and see which ones are moving into better/worse environments. A human can do a handful, but if you have 12 supercells out there - the PC will win.

Actually those private forecasters are using models that people outside of their environment don't have access to. It's cheaper to pay AW $100K per year than it is to develop a $100M modeling system.

I appreciate your concern for the plight of the human forecaster but I'm just not sure you are up to speed on where things are now (and what that means for the future.)

I'm completely aware of warn-on-forecast. To get to the level of human forecasters, it requires having very accurate model data many minutes/hours before storms form. Again, if one storm is off by 10 miles, you could be tornado warning a population of 1,000 vs. a population of 50,000 in some cases.

You're telling me that a model will be able to analyze the live radar and release warnings immediately? I would love to see a live, minute-by-minute, scan-by-scan model that's constantly updating with up-to-the-minute data. I know that's ultimately the goal of warn-on-forecast, but I have my doubts about how well the models will do with accurate detection (both location and intensity) and the false alarm rate. At WeatherBug, we have DTAs (Dangerous Thunderstorm Alerts) that are automatic warnings for storms based on live radar and lightning, but it certainly has its limitations.

What models are we (private forecasters) using that others don't have access to? The Euro? You can get that data for as little as $15-20/month from places like WeatherBell and StormVista. Some companies (Like MDA Weather Services/EarthSat, which I used to work for) have other Euro data available at a higher premium, but even those higher rates would be cheaper than hiring an actual met. MDA has a proprietary "superensemble" that combines select model data and automatically weighs each one, but even then, the human forecasters that work there can beat this proprietary information. About 95% of the forecasting tools I use at my private forecasting job are free to the public.

I live in the private forecasting world, and try to keep up with new forecasting model developments as much as I can. I even took classes on NWP and atmospheric modeling when I was in school. I know where the model limitations are, and can exploit them accordingly to put out a better forecast. That is my knowledge base. Do I know every little detail about all of the latest research? No. I'm not directly involved with them. Do I know what they're trying to accomplish, and where some of the flaws are? Certainly.

 

[Todd Lemery]

It pretty much boils down to this. No matter where we are right now with computers putting out short or long term forecasts, the computer will eventually surpass the human mind for comprehensive and critical thought. Unless we figure out a way to greatly increase a human's ability to think, the computer will eventually take over many functions in our lives. The cell phone in our pockets has a lot more computing power than the high tech(at the time) computers that helped land us on the moon.

 

[Marcus Diaz]

Side note...this thread should be moved to Advanced Weather instead of Winter Weather.

 

[rdale]

Mark - private sector companies use models they develop internally, or take open source products (like the WRF) and run them through extensive tweaking to accomplish forecast skills that fit their clients' needs. To get to that point is a little more costly than $20 a month

 

[Mark Ellinwood]

You do have a point... some companies do run their own models that are specifically tailored for their clients. But if they're anything like the RPM, they may be causing more problems than solutions Like I said though, the vast majority of forecasting data that private forecasters use are available for free or very cheaply. Do the tailored models eliminate meteorologists (meaning forecasters) completely, as is suggested by the thread topic? No. As I said before, as computers get more accurate, meteorologists will still find other ways to improve upon the existing forecast, if not in the more conventional ways that we're used to. It won't take as many human forecasters to cover all the work in the future (as programmers become more prominent), but there will still be human forecasters.

Heck, there could even be other ways to make an automated forecast that we haven't even thought about yet, and there could be new methods for human forecasters to beat computer forecasts that have yet to be discovered. Our field is still relatively young, and I think there's still huge gains to be made both by humans and by forecasters in the years to come. I guess my point since I started posting in this thread is to not think of human forecasters as static while the world of computing remains dynamic. We learn as the computers learn. We evolve as the computers evolve. Will certain areas of forecasting be replaced by computers eventually? Yes. Will forecasting as a profession go away as a result? No. Forecasters will just find other things to forecast and be better at.

Also, just look at it from a practical standpoint. In the 80s and 90s, the models gained a bunch of skill very quickly as the knowledge of the weather from a mathematical standpoint grew with the better technology. Since about the early 2000s, notable increase in skill has been hard to come by (at least looking at the more traditional models like the GFS, NAM, ECMWF, UKMET, CMC, etc.) as the mathematical discoveries and revelations that helped create significantly better models waned. More and more we are relying on leaps in technology than we are in leaps in meteorology in order to advance the model skill. How much could be debated, but it's clear that the trend for increased skill is much slower than it was just a decade or two ago. Finer resolution models offer higher skill in select areas, but suffer in other areas. There's always a trade-off.

Computers are limited by spatial and temporal resolution, availability and accuracy of observations (in the horizontal and vertical), what equations and assumptions are used, rounding errors, compounding errors as you extend out into time, computing power, etc., etc. Each of these are MAJOR hurdles to overcome to the point where computers consistently out-forecast humans. With all of these ways for a computer to fail, you can see why I'm very skeptical about computers completely taking over the forecasting field.

 

[Kevin Rimcoski]

You do have a point... some companies do run their own models that are specifically tailored for their clients. But if they're anything like the RPM, they may be causing more problems than solutions Like I said though, the vast majority of forecasting data that private forecasters use are available for free or very cheaply. Do the tailored models eliminate meteorologists (meaning forecasters) completely, as is suggested by the thread topic? No. As I said before, as computers get more accurate, meteorologists will still find other ways to improve upon the existing forecast, if not in the more conventional ways that we're used to. It won't take as many human forecasters to cover all the work in the future (as programmers become more prominent), but there will still be human forecasters.

Heck, there could even be other ways to make an automated forecast that we haven't even thought about yet, and there could be new methods for human forecasters to beat computer forecasts that have yet to be discovered. Our field is still relatively young, and I think there's still huge gains to be made both by humans and by forecasters in the years to come. I guess my point since I started posting in this thread is to not think of human forecasters as static while the world of computing remains dynamic. We learn as the computers learn. We evolve as the computers evolve. Will certain areas of forecasting be replaced by computers eventually? Yes. Will forecasting as a profession go away as a result? No. Forecasters will just find other things to forecast and be better at.

Also, just look at it from a practical standpoint. In the 80s and 90s, the models gained a bunch of skill very quickly as the knowledge of the weather from a mathematical standpoint grew with the better technology. Since about the early 2000s, notable increase in skill has been hard to come by (at least looking at the more traditional models like the GFS, NAM, ECMWF, UKMET, CMC, etc.) as the mathematical discoveries and revelations that helped create significantly better models waned. More and more we are relying on leaps in technology than we are in leaps in meteorology in order to advance the model skill. How much could be debated, but it's clear that the trend for increased skill is much slower than it was just a decade or two ago. Finer resolution models offer higher skill in select areas, but suffer in other areas. There's always a trade-off.

Computers are limited by spatial and temporal resolution, availability and accuracy of observations (in the horizontal and vertical), what equations and assumptions are used, rounding errors, compounding errors as you extend out into time, computing power, etc., etc. Each of these are MAJOR hurdles to overcome to the point where computers consistently out-forecast humans. With all of these ways for a computer to fail, you can see why I'm very skeptical about computers completely taking over the forecasting field.

This pretty much covers it seems like. There's still going to be a need for human forecasters, as well as NWP models. Both are going to have limitations and strengths obviously but neither are going to be any good without the other.
Say you're out chasing a storm...based on environmental conditions like inflow/outflow Temps you can make an adjustment on your now cast that a computer might not be able to pick up on, whether the scale is too small, or the event is too quick and the computer model hasn't updated yet

 

[rdale]

nsx: If that data is being relayed back to the meteorologist at ABCWeather.com - can't the computer at ABCWeather incorporate it too? In 50 years every car in the US will be sending realtime weather conditions -- I see no way that a human plotting that data and modifying her nowcast will do so faster than the black box forecaster.

 

[Kevin Rimcoski]

nsx: If that data is being relayed back to the meteorologist at ABCWeather.com - can't the computer at ABCWeather incorporate it too? In 50 years every car in the US will be sending realtime weather conditions -- I see no way that a human plotting that data and modifying her nowcast will do so faster than the black box forecaster.

And you come in and smack me down haha just kidding.
That's a good point, and a bit scary to think about

 

[rdale]

It's not a smack This is all just prediction... Honestly though I wouldn't be "scared" of it - I would LOVE to know that a mile ahead of me people are pulling off the road because of hail, or the turn I'm about to take has icy patches that have caused drivers to lose traction briefly, etc. Will I be sad that I'm chasing a storm based on a computer projection that's better than my personal forecast? No, not really... I'll be happier to see the big one regardless of how I get there. Or I can go "old school" if I really wanted to...

I had a similar discussion in another arena the other day. I would "romanticize" about sitting on the back porch with my GE SuperRadio in 1994 listening to Yankees games. Now I have the games on my phone and can even watch live TV if I want. Seems so 'impersonal' -- but you know what? Most of the time I had to move the antenna around to get a good signal from 400 miles away, and daytime it was useless. If I had to do something I couldn't lug the radio around with me. If I had to hit the bathroom, I had to hit the pause button, etc. I think weather will be the same way. Yes we'll look back on "reflect" on the days when we had to do it manually, but even in chasing I never want to go back to the days when I all had was my ham radio talking to net control who was relaying his interpretation of the TWC radar display...

 

[StephenHenry]

Posting in this old thread that Calvin started in honor of the VA Enhanced Severe Outlook today

Ars Technica ran an article on this exact topic today. I've always thought they do a nice job on tech coverage, and this seemed pretty decent for a venue not dedicated to weather. Mostly it reinforced how much I like James Spann.

http://arstechnica.com/science/2016...-models-do-we-still-need-weather-forecasters/

 

[JamesCaruso]

Nice to see this thread revived, I must have missed it the first time...

I am not a meteorologist so perhaps I am not qualified to weigh in, but personally I think the answer to the question in the title of this thread is NO. The specific jobs and careers in meteorology will drastically CHANGE like every other field affected by technology, but meteorologists will not become extinct.

I look at my own field - accounting - as an example. Technology has clearly automated much of what accountants used to do with pencils and green ledger pads, yet accountants are more in demand than ever. Computers have resulted in more data to be analyzed, more information to draw insights from, more complexity to be interpreted in a way that is simple enough for non-financial people to understand and use for decision-making. Accountants do less of the processing but more analysis and provide interpretive insight to individuals and business executives. And there are totally new careers blending technology and accounting: designing systems, configuring/implementing software and automated processes for companies, etc.

I see the roles of meteorologists changing in this way. Hasn't the overall weather enterprise grown so far as a result of technology? Private weather companies didn't even exist before technology enabled providing hyper-local forecasts and warnings customized for individual businesses. So yeah maybe certain operational forecasting roles become extinct but the field itself is going to change, not disappear, and there will continue to be jobs in it, including ones we can't even envision today. I remember seeing an economist presentation where he said that some very high percentage (I forget the exact number) of jobs listed in the Bureau of Labor statistics did not even exist a few decades ago.


Sent from my iPad using Tapatalk