HPC

[Chagrin Falls]

After the ECMWF nailed the system this week the best the HPC learned is to take the middle ground between the extremes?

EXTENDED FORECAST DISCUSSION NWS HYDROMETEOROLOGICAL PREDICTION CENTER CAMP SPRINGS MD 1238 PM EST MON DEC 13 2010 VALID 12Z THU DEC 16 2010 - 12Z MON DEC 20 2010 ...PRELIMINARY UPDATE... USED THE 00Z ECMWF TO UPDATE THE FRONTS AND PRESSURES FOR DAYS 3 AND 4...WITH A GRADUAL TRANSITION TO THE 06Z GEFS MEAN BY DAY 6. THE BLOCKY REGIME IS INDICATED TO HOLD THROUGH THE PERIOD BY ALL THE MODELS...WITH A SUPPRESSED POLAR JET ACROSS THE ENTIRE NATION. HEIGHTS WILL BE THE GREATEST OVER THE SOUTHERN HIGH PLAINS WHERE THE SUBTROPICAL RIDGE WILL BE ABLE TO HERNIATE NORTHWARD. THE DETAILS OF THE ENERGY COMING INTO THE WEST COAST ARE WILDLY UNCERTAIN...BUT THERE IS ENOUGH AGREEMENT BETWEEN THE ENSEMBLE MEANS FROM THE VARIOUS MODELING CENTERS TO SUPPORT THE BLEND USED FOR THIS FORECAST PACKAGE. THE SITUATION ALONG THE EAST COAST IS FAR MORE CONTENTIOUS...WITH THE 00Z AND 06Z GFS RUNS BOTH BRINGING A MAJOR SNOWSTORM TO THE MID ATLANTIC AND NORTHEAST DAYS 6 AND 7. THE 00Z ECMWF KEEPS THE NORTHERN STREAM DOMINANT OVER THIS REGION...AS DOES THE GEM GLOBAL. HOWEVER...THE 12Z/12 AND 12Z/11 RUNS OF THE ECMWF BOTH SHOWED THE SNOWSTORM. OPTED TO GO WITH THE 06Z GEFS MEAN FOR THIS POTENTIAL EVENT...WHICH AT LEAST BRINGS A CYCLONE NORTHWARD OVER THE WESTERN ATLANTIC...JUST FARTHER OFFSHORE AND WITH LESS ISOBARIC COMMITMENT...IF YOU WILL. THIS CHOICE LEAVES ROOM FOR TRENDING EITHER WAY. ...FINAL... THE 12Z GUIDANCE OFFERS A DISCONCERTING ARRAY OF POSSIBLE OUTCOMES FOR THE SPLIT FLOW OVER THE NATION DURING THIS MEDIUM RANGE PERIOD. FORTUNATELY...THE LATEST GEFS MEAN HAS NOT TRENDED MUCH FROM THE 06Z RUN...WHICH INFORMED MUCH OF THE UPDATE PACKAGE. THE DETERMINISTIC GFS HAS GONE FLATTER WITH THE ATLANTIC COAST WAVE DAY 6...WITH THE GEFS MEAN SHOWING MORE AMPLITUDE...ENOUGH TO THREATEN PORTIONS OF THE EASTERN SEABOARD WITH SNOW. THE GEM GLOBAL HAS TRENDED CONSIDERABLY MORE AMPLIFIED WITH THE DAY 6 SYSTEM...BRINGING SNOW BACK TO INTERSTATE 95 FROM RICHMOND NORTHWARD. THE UKMET REMAINS ON ITS OWN WITH SHOWING ENOUGH AMPLITUDE TO BACK THE FLOW OVER THE ATLANTIC STATES DAYS 5 AND 6 TO BRING RAIN TO MUCH OF THE REGION. THE LESSON WITH THIS MOST RECENT COMPLEX AMPLIFICATION OVER THE MIDWEST AND EAST WAS TO GO BETWEEN THE EXTREME SOLUTIONS...WHICH THE UPDATE BLEND ACCOMPLISHED. FOR THIS REASON...NO CHANGES WERE MADE FOR THE FINAL ISSUANCE. CISCO

Below are the ECMWF ensemble mean, ECMWF operational run and GFS operational run

96 hr progs initalized last Wednesday @ 12Z. Output is for Sunday @ 12 Z. Additionally,

I posted the ECMWF initalization for 12Z Sunday.

The ECMWF was, what, at most 50 miles off the actual center low position? For the GFS, let's see, how far is

Greensboro, NC from Toledo?

I don't think I've ever seen NWS revisionist history like this before. Seems to me the "extreme" solution worked out just dandy, as those that didn't buy the outlier busted (ask most of the NWS WFO's in the Mid-west).

[usedtobe]

I don't view that as revisionist history. I think they forecast the low too far east and feel they would have done better to go with a more middle of the road solution last time, that would have allowed them to trend towards the correct solution. For sure, the Euro was way better which is often the case with a strongly amplifying pattern with no block nearby. However, it's not always the winner even though it on average scores better. More often than not, the best choice in the longer ranges when no models have that much skill is to go with the middle of the road solution and then trend towards the ultimate one. If they had completely bought the ECMWF solution a day ago for the dec 19 storm they would have been forced to back off considerable over the past 24 hrs. Spltting the difference is sort of like going with an ensemble mean during a tough forecast. You know it probably won't be right but are trying to minimize your losses. I guess the bottom line is I don't think Cisco was trying to be pull the wool over your eyes.

[Ed Lizard]

I don't view that as revisionist history. I think they forecast the low too far east and feel they would have done better to go with a more middle of the road solution last time, that would have allowed them to trend towards the correct solution. For sure, the Euro was way better which is often the case with a strongly amplifying pattern with no block nearby. However, it's not always the winner even though it on average scores better. More often than not, the best choice in the longer ranges when no models have that much skill is to go with the middle of the road solution and then trend towards the ultimate one. If they had completely bought the ECMWF solution a day ago for the dec 19 storm they would have been forced to back off considerable over the past 24 hrs. Spltting the difference is sort of like going with an ensemble mean during a tough forecast. You know it probably won't be right but are trying to minimize your losses. I guess the bottom line is I don't think Cisco was trying to be pull the wool over your eyes.

We need a visit from the Great D R

[Srain]

We need a visit from the Great D R

He has been rather active today in other threads. Just sayin...

[Guest someguy]

CHAGRIN

I did a post on this over this past weekend

mentioned the very same points

see the pinned WOOF thread page 4 Post # 60 or #61

[Chagrin Falls]

I don't view that as revisionist history. I think they forecast the low too far east and feel they would have done better to go with a more middle of the road solution last time, that would have allowed them to trend towards the correct solution. For sure, the Euro was way better which is often the case with a strongly amplifying pattern with no block nearby. However, it's not always the winner even though it on average scores better. More often than not, the best choice in the longer ranges when no models have that much skill is to go with the middle of the road solution and then trend towards the ultimate one. If they had completely bought the ECMWF solution a day ago for the dec 19 storm they would have been forced to back off considerable over the past 24 hrs. Spltting the difference is sort of like going with an ensemble mean during a tough forecast. You know it probably won't be right but are trying to minimize your losses. I guess the bottom line is I don't think Cisco was trying to be pull the wool over your eyes.

On one hand I understand what you are saying. But on the other hand, I must respectfully disagree.

First of all, yes in many cases ensemble means or consensus output is the most reasonable solution, especially if their are initialization problems or if known model biases are evident.

However, I'm a bit troubled by your response. This is nothing personal, but I often wonder if the product meteorologists put out has any value (this thought includes my forecasts, by the way) especially in the medium range.

I know this is going to rub people the wrong way, but I've got to say this because it is something I can't understand. More often than not, WFOs and HPC discussions take a consensus, blend or averaging of several numerical models. Instead of putting emphasis on the one that is statistically more accurate (in this case the ECMWF ensemble means), they usually just average them all together. I've seen this for at least the last 20 years. Heck, in the 1990s it was hard to get the NWS to admit the ECMWF existed, even though it was FAR superior to the MRF and AVN time and time again.

But if all meteorologists do most of the time is average/blend or make a model consensus, there is NO scientific/meteorological analysis and reasoning. All meteorologists seem to do is a "split-the-difference" mechanical approach that averages opinions of unequal quality and skill (various model output).

If this is the case, why have a meteorologist? I can create a program on an old TANDY computer that would make a consensus of the various global model output. And the TANDY doesn't need a pension, medical coverage or sick days.

A meteorologist is present for meteorological reasoning. A computer can make a consensus.

And if in hindsight the only thing that is learned is a reaffirm a product that can never be superior to the model consensus itself, then why bother?

I think if meteorologists continue to use model consensus for their forecasts they will outsource their products to 100% automation sooner than needed. Rarely are they adding value to something a computer can do much more quickly and objectively. I don't think this should be the case.

[Ed Lizard]

He has been rather active today in other threads. Just sayin...

Under the same name? I used to e-mail him before I knew he was on a forum when the HPC final disco would suggest marked disagreement with the NHC/TPC handling of features.

[usedtobe]

On one hand I understand what you are saying. But on the other hand, I must respectfully disagree.

First of all, yes in many cases ensemble means or consensus output is the most reasonable solution, especially if their are initialization problems or if known model biases are evident.

However, I'm a bit troubled by your response. This is nothing personal, but I often wonder if the product meteorologists put out has any value (this thought includes my forecasts, by the way) especially in the medium range.

I know this is going to rub people the wrong way, but I've got to say this because it is something I can't understand. More often than not, WFOs and HPC discussions take a consensus, blend or averaging of several numerical models. Instead of putting emphasis on the one that is statistically more accurate (in this case the ECMWF ensemble means), they usually just average them all together. I've seen this for at least the last 20 years. Heck, in the 1990s it was hard to get the NWS to admit the ECMWF existed, even though it was FAR superior to the MRF and AVN time and time again.

But if all meteorologists do most of the time is average/blend or make a model consensus, there is NO scientific/meteorological analysis and reasoning. All meteorologists seem to do is a "split-the-difference" mechanical approach that averages opinions of unequal quality and skill (various model output).

If this is the case, why have a meteorologist? I can create a program on an old TANDY computer that would make a consensus of the various global model output. And the TANDY doesn't need a pension, medical coverage or sick days.

A meteorologist is present for meteorological reasoning. A computer can make a consensus.

And if in hindsight the only thing that is learned is a reaffirm a product that can never be superior to the model consensus itself, then why bother?

I think if meteorologists continue to use model consensus for their forecasts they will outsource their products to 100% automation sooner than needed. Rarely are they adding value to something a computer can do much more quickly and objectively. I don't think this should be the case.

Your right about if you always take consensus you could easily write a program to do the job. I think the met probably was responding to the earlier forecast. It's hard not to. I've had times when I really backed off on making significant changes to a model after i got my brains beat in when the forecast was verified. My comment were mostly about your implications about his motives. For this coming storm, what would you do, if you used the Euro you would have waffled back and forth between dc getting lots of snow or none depending on the run. Now the GFS has tilted back towards the heavier snow solution. In reality, our forecasts should be couched in probabilities, then you don't have to make a pure compromise. The trouble with the MN snowstorm was it was probably going to be one solution or the other. Anyway, we can agree to disagree.

[baroclinic_instability]

On one hand I understand what you are saying. But on the other hand, I must respectfully disagree.

First of all, yes in many cases ensemble means or consensus output is the most reasonable solution, especially if their are initialization problems or if known model biases are evident.

However, I'm a bit troubled by your response. This is nothing personal, but I often wonder if the product meteorologists put out has any value (this thought includes my forecasts, by the way) especially in the medium range.

I know this is going to rub people the wrong way, but I've got to say this because it is something I can't understand. More often than not, WFOs and HPC discussions take a consensus, blend or averaging of several numerical models. Instead of putting emphasis on the one that is statistically more accurate (in this case the ECMWF ensemble means), they usually just average them all together. I've seen this for at least the last 20 years. Heck, in the 1990s it was hard to get the NWS to admit the ECMWF existed, even though it was FAR superior to the MRF and AVN time and time again.

But if all meteorologists do most of the time is average/blend or make a model consensus, there is NO scientific/meteorological analysis and reasoning. All meteorologists seem to do is a "split-the-difference" mechanical approach that averages opinions of unequal quality and skill (various model output).

If this is the case, why have a meteorologist? I can create a program on an old TANDY computer that would make a consensus of the various global model output. And the TANDY doesn't need a pension, medical coverage or sick days.

A meteorologist is present for meteorological reasoning. A computer can make a consensus.

And if in hindsight the only thing that is learned is a reaffirm a product that can never be superior to the model consensus itself, then why bother?

I think if meteorologists continue to use model consensus for their forecasts they will outsource their products to 100% automation sooner than needed. Rarely are they adding value to something a computer can do much more quickly and objectively. I don't think this should be the case.

You make some valid points, and you sort of answer your own question. Why are there meteorologists? Because we still need to interpret the data. Look at the NWS. Read the 2020 strategy guide. What are they doing? Trending more towards an "impact based" forecasting that has been going on a long time in private weather. Both short-range weather (within 2-3 days) and weather impacts in general require a good meteorologist. No meteorologist wants to admit it, but the models can generally beat the meteorologists and there isn't necessarily a large need to put much effort into the forecast 5-7 days out. The real effort comes in the short-range where weather impacts can be disseminated, meteorologists can work with emergency managers on potential impacts, and weather forecasts can be tailored more for potential societal impacts. This is where both private and public weather forecasting is trending. The bad news is computers are in some ways taking away jobs; the good news is talented meteorologists who can effectively communicate with the public/customer will always be in need and will never be replaced.

[baroclinic_instability]

To continue with those thoughts, what do I mean by "impact based"? In the private sector, a very good example is road weather meteorology/forecasting where the forecast is tailored specifically for the road management /state DOT community. What do they care about? Road weather potential. The main goal of the private road weather community is to forecast the potential for any road snow/ice accumulations so they can properly implement a plan to anti-ice since it is FAR more cost-effective to keep roads free of ice as opposed to de-icing the roads. Moreover, they are quite interested in the overall traffic impacts. Will snow come down hard for 4 hours at 1+ inch per hour? Or will it be 4 inches over a 12 hour period? Will the snow fall during rush-hour or overnight on a weekend? Road weather managers don't care if snow falls and accumulates in the grass but melts instantly because roads are too warm. All these things matter and they have significant societal impacts. As for the NWS, they are also trending towards these types of forecasts. Apparently, if I have heard correctly, some offices are already testing "impact-based" forecasting with the weather products they issue. For instance, they may issue a winter weather advisory for 3" snow at rush hour, but they won't if it falls overnight. Either way, impact based forecasting requires more effort (in my opinion) since both weather/impacts need to be assessed so as to limit the societal impacts. Pf course this has always been why weather organizations have been around, but it is being far more aggressively approached as computing power automates a lot of tasks.

[Guest someguy]

Both short-range weather (within 2-3 days) and weather impacts in general require a good meteorologist. No meteorologist wants to admit it, but the models can generally beat the meteorologists and there isn't necessarily a large need to put much effort into the forecast 5-7 days out

WHAT?

you dont work in the energy or AG fields do you?

[baroclinic_instability]

WHAT?

you dont work in the energy or AG fields do you?

I am taking it from the perspective he was using which is operational public forecasting and I added private wx as well. Under that scenario, the time needed to put together good long range discussions/forecasts is not justified at all. The cost/benefit is just not there when far more valuable tasks can be accomplished in that same time. We can indeed beat the models, but not by a lot in the 5-7 day range...and it takes quite a bit of additional time for a somewhat small benefit. This potential noreaster is a good example.

Longer range trends for use in energy or AG is a completely different discussion.

[WeatherQ]

I am taking it from the perspective he was using which is operational public forecasting and I added private wx as well. Under that scenario, the time needed to put together good long range discussions/forecasts is not justified at all. The cost/benefit is just not there when far more valuable tasks can be accomplished in that same time. We can indeed beat the models, but not by a lot in the 5-7 day range...and it takes quite a bit of additional time for a somewhat small benefit. This potential noreaster is a good example.

Longer range trends for use in energy or AG is a completely different discussion.

Meteorology is just like any scientific endeavor that uses historic outcomes, current data and observation to hypothesize upon a proposed outcome or expected result. I do enjoy when the threat board lights up and the mets, pro forecasters and avg met enthusiasts (like myself) sit back and watch as the real dealers go about their craft comparing historic analog forecasts with predicted future outcomes (forecasts) demonstrating the correlation between historic trending and fresh data interpretation. The great thing about meteorology (IMHumbleO) is that some of the variables do not always lend themselves to predictability. During my private pilot training, I often (and thankfully only scarily once) encountered the unpredictability of weather.

[Guest someguy]

I am taking it from the perspective he was using which is operational public forecasting and I added private wx as well. Under that scenario, the time needed to put together good long range discussions/forecasts is not justified at all. The cost/benefit is just not there when far more valuable tasks can be accomplished in that same time. We can indeed beat the models, but not by a lot in the 5-7 day range...and it takes quite a bit of additional time for a somewhat small benefit. This potential noreaster is a good example.

Longer range trends for use in energy or AG is a completely different discussion.

well as long as you make the distinction .....

[winterymix]

More often than not, WFOs and HPC discussions take a consensus, blend or averaging of several numerical models. Instead of putting emphasis on the one that is statistically more accurate (in this case the ECMWF ensemble means), they usually just average them all together.

Experience counts for something.

Here is a portion of the AFD from Sterling:

LONG TERM /FRIDAY THROUGH MONDAY/...

-- Changed Discussion --EXTENDED PERIOD IS QUITE UNCERTAIN WITH MODEL OUTPUT CONTINUING TO

FLIP ON A RUN-TO-RUN BASIS. WITH THE UNCERTAINTY...LITTLE CHANGE WAS

MADE WITH THIS FORECAST PACKAGE. NEW ECMWF HAS ALMOST NOTHING OVER

THE AREA ALL THE WAY THROUGH THE EXTENDED...LATEST GFS SPREADS A

SWATH OF PRECIP ACROSS THE AREA ON SATURDAY NIGHT/EARLY SUNDAY.

A meteorologist that has been forecasting for the region of interest for a number of winters and has

seen similar situations and has a familiarity with model basis brings a lot of value to creating a

forecast of least regret. Why criticize a meteorologist for creating a weighted average of two

independently derived models? A weighted average implies that the meteorologist is already down-playing model output that is suspect.

[MGorse]

To continue with those thoughts, what do I mean by "impact based"? In the private sector, a very good example is road weather meteorology/forecasting where the forecast is tailored specifically for the road management /state DOT community. What do they care about? Road weather potential. The main goal of the private road weather community is to forecast the potential for any road snow/ice accumulations so they can properly implement a plan to anti-ice since it is FAR more cost-effective to keep roads free of ice as opposed to de-icing the roads. Moreover, they are quite interested in the overall traffic impacts. Will snow come down hard for 4 hours at 1+ inch per hour? Or will it be 4 inches over a 12 hour period? Will the snow fall during rush-hour or overnight on a weekend? Road weather managers don't care if snow falls and accumulates in the grass but melts instantly because roads are too warm. All these things matter and they have significant societal impacts. As for the NWS, they are also trending towards these types of forecasts. Apparently, if I have heard correctly, some offices are already testing "impact-based" forecasting with the weather products they issue. For instance, they may issue a winter weather advisory for 3" snow at rush hour, but they won't if it falls overnight. Either way, impact based forecasting requires more effort (in my opinion) since both weather/impacts need to be assessed so as to limit the societal impacts. Pf course this has always been why weather organizations have been around, but it is being far more aggressively approached as computing power automates a lot of tasks.

This is correct. There has been much more emphasis in the NWS the last several years on more near/short term impact-based forecasting. A good example of this is the winter weather advisory issued early this morning by NWS Sterling, VA.

[Analog96]

I am taking it from the perspective he was using which is operational public forecasting and I added private wx as well. Under that scenario, the time needed to put together good long range discussions/forecasts is not justified at all. The cost/benefit is just not there when far more valuable tasks can be accomplished in that same time. We can indeed beat the models, but not by a lot in the 5-7 day range...and it takes quite a bit of additional time for a somewhat small benefit. This potential noreaster is a good example.

Longer range trends for use in energy or AG is a completely different discussion.

Ok BUT...

what if the first three days of your forecast period are sunny and an easy forecast and day five is where the big storminess starts?

[Chagrin Falls]

Experience counts for something.

Here is a portion of the AFD from Sterling:

A meteorologist that has been forecasting for the region of interest for a number of winters and has

seen similar situations and has a familiarity with model basis brings a lot of value to creating a

forecast of least regret. Why criticize a meteorologist for creating a weighted average of two

independently derived models? A weighted average implies that the meteorologist is already down-playing model output that is suspect.

Hello,

My intent wasn't to criticize a specific meteorologist. I'm attempting to create a discussion on the pro's/con's of "model consensus". Obviously, I often see it as a negative with little value to forecast users, while some see it differently.

As computers become more powerful and numerical modeling becomes more accurate I think the job of a operational meteorologist at the very least becomes more specialized (see the above posts by other mets). Sometimes, I wonder if meteorologists are slowly becoming outsourced. Even with specialized forecasts such as aviation (I use these products a lot) I find little value in the TAF vs. the automated LAMP MOS. In fact, sometimes I prefer the LAMP MOS because of the 1 hr time interval.

I went to school in the 1990s. In our forecast discussions we were taught to do a thorough analysis, recognize specific long-wave pattern traits, teleconnections, etc., and then look at various model output and use experience and our scientific knowledge to determine which model had a more reasonable output. Obviously, models have become more accurate since the 1990s and ensemble forecasting has really come a long way. Forecasts are more accurate because numerical modeling has advanced.

Computers don't have experience of a seasoned forecaster and they can't recognize patterns. Instead of blending model output most of the time (which can be done very easily by a computer for little cost), I guess I'd rather see a meteorologist at least attempt to do a through analysis and then pick a model that he thinks has the better solution. To me, model consensus seems a bit like the days of "hugging MOS".

In closing, I'm a bit concerned that some day numerical modeling will be so advanced that humans will not be able to add anything of value. The only jobs left for meteorologists would be those that do research, those that design numerical modeling and those that fix ASOS/radar systems.

[baroclinic_instability]

Hello,

My intent wasn't to criticize a specific meteorologist. I'm attempting to create a discussion on the pro's/con's of "model consensus". Obviously, I often see it as a negative with little value to forecast users, while some see it differently.

As computers become more powerful and numerical modeling becomes more accurate I think the job of a operational meteorologist at the very least becomes more specialized (see the above posts by other mets). Sometimes, I wonder if meteorologists are slowly becoming outsourced. Even with specialized forecasts such as aviation (I use these products a lot) I find little value in the TAF vs. the automated LAMP MOS. In fact, sometimes I prefer the LAMP MOS because of the 1 hr time interval.

I went to school in the 1990s. In our forecast discussions we were taught to do a thorough analysis, recognize specific long-wave pattern traits, teleconnections, etc., and then look at various model output and use experience and our scientific knowledge to determine which model had a more reasonable output. Obviously, models have become more accurate since the 1990s and ensemble forecasting has really come a long way. Forecasts are more accurate because numerical modeling has advanced.

Computers don't have experience of a seasoned forecaster and they can't recognize patterns. Instead of blending model output most of the time (which can be done very easily by a computer for little cost), I guess I'd rather see a meteorologist at least attempt to do a through analysis and then pick a model that he thinks has the better solution. To me, model consensus seems a bit like the days of "hugging MOS".

In closing, I'm a bit concerned that some day numerical modeling will be so advanced that humans will not be able to add anything of value. The only jobs left for meteorologists would be those that do research, those that design numerical modeling and those that fix ASOS/radar systems.

I don't know why you say this, especially with the points I brought up in the previous points. There will always be a need fore meteorologists to both interpret and disseminate the information. It does take talent to take in all this scientific data and create impact based forecasts which are easily understandable to non-scientific users. There will always be weather forecasters and short-term weather is a great example where I don't see automation taking over anytime soon.

[Cory]

Computers don't have experience of a seasoned forecaster and they can't recognize patterns. Instead of blending model output most of the time (which can be done very easily by a computer for little cost), I guess I'd rather see a meteorologist at least attempt to do a through analysis and then pick a model that he thinks has the better solution. To me, model consensus seems a bit like the days of "hugging MOS".

Artificial Neural Networks are capable of "thinking" and recognizing patterns like a seasoned human forecaster. It's already technically possible to have a computer analyze output from various models, recognize differences and account for model biases, compare the situation to past situations and current conditions, make forecast decisions, and learn from mistakes. Luckily (jobwise) not as much effort/funds have been put behind neural networks as compared to deterministic modeling and other technologies.

[baroclinic_instability]

Artificial Neural Networks are capable of "thinking" and recognizing patterns like a seasoned human forecaster. It's already technically possible to have a computer analyze output from various models, recognize differences and account for model biases, compare the situation to past situations and current conditions, make forecast decisions, and learn from mistakes. Luckily (jobwise) not as much effort/funds have been put behind neural networks as compared to deterministic modeling and other technologies.

It still brings up the same point in that whatever way you look at it, pure "forecasting" jobs are definitely declining with increasing computing power/technology. The job of traditional forecasters is just changing.

[uncle W]

A question.....

Should the extended forecast read Snow or rain on Sunday if snow is the greatest possibility or rain or snow if rain is the greatest possibility?...Or it doesn't matter?...I would use snow or rain if snow was the greatest chance and rain or snow if I thought rain was the greatest possibility...It would show where a forecaster was leaning...Now if the chances were even in my mind I'd flip a coin...

[Analog96]

A question.....

Should the extended forecast read Snow or rain on Sunday if snow is the greatest possibility or rain or snow if rain is the greatest possibility?...Or it doesn't matter?...I would use snow or rain if snow was the greatest chance and rain or snow if I thought rain was the greatest possibility...It would show where a forecaster was leaning...Now if the chances were even in my mind I'd flip a coin...

Normally, snow comes first if snow is more probable. Personally, for the weekend, I would go with just snow for now, but not excessively high chances, since there is a good deal of spread, and out to sea is probably more likely than rain right now.

[bsperlin]

Normally, snow comes first if snow is more probable. Personally, for the weekend, I would go with just snow for now, but not excessively high chances, since there is a good deal of spread, and out to sea is probably more likely than rain right now.

This is a discussion I've had with Meteorologists and no one's had a great insight (including me). How about in the above mentioned case: "Probably Snow with a possibility of rain" and the reverse: "Probably Rain with a possibility of snow". Also note the capitalizations on the word after "Probably".

[bsperlin]

Artificial Neural Networks are capable of "thinking" and recognizing patterns like a seasoned human forecaster. It's already technically possible to have a computer analyze output from various models, recognize differences and account for model biases, compare the situation to past situations and current conditions, make forecast decisions, and learn from mistakes. Luckily (jobwise) not as much effort/funds have been put behind neural networks as compared to deterministic modeling and other technologies.

I expect this to be a trend, though I don't think AI will advance to the point of totally eliminating meteorologists in my lifetime (I'm old, however). When I started working there were "secretarial pools". They disappeared when middle managers were given computers and told to type their own damn letters. There also was a thriving steel and auto industry. While a lot of that has been lost as a result of overseas out-sourcing, the fact is that foreign (and domestic) plants of both types have become highly automated/robotic and many have lost their jobs. The upper midwest is just holding on by its fingernails. So, yes, the barrel of the gun IS aimed at meteorologists, but the bullet is quite slow as AI improves like molasses moves.