New 25-Day Forecast on AccuWeather

[RodneyS]

See http://www.accuweath...-forecast/63449

This raises two questions in my mind: (1) How likely is it that AccuWeather's forecasts beyond 15 days will beat random chance? (2) Will AccuWeather or any third party be tracking the accuracy of these forecasts?

[donsutherland1]

See http://www.accuweath...-forecast/63449

This raises two questions in my mind: (1) How likely is it that AccuWeather's forecasts beyond 15 days will beat random chance? (2) Will AccuWeather or any third party be tracking the accuracy of these forecasts?

I didn't find any actual 25-day forecasts on the site. Once the 25-day forecasts start being released, it might be interesting to track Day 20 and Day 25 verification for a time and compare it with climatology.

[Jesse Ferrell - AccuWx]

I didn't find any actual 25-day forecasts on the site. Once the 25-day forecasts start being released, it might be interesting to track Day 20 and Day 25 verification for a time and compare it with climatology.

It's there, click on "Month"

We'll certainly be tracking the accuracy internally as we do with our 15-day forecast.

[donsutherland1]

It's there, click on "Month"

We'll certainly be tracking the accuracy internally as we do with our 15-day forecast.

Many thanks. I also wish AccuWx success with the 25-day forecasts. Hopefully, the audience seeing them properly understands that verification cannot be anything close to the near-term forecasts due to the much larger degree of uncertainty involved.

[Jesse Ferrell - AccuWx]

Many thanks. I also wish AccuWx success with the 25-day forecasts. Hopefully, the audience seeing them properly understands that verification cannot be anything close to the near-term forecasts due to the much larger degree of uncertainty involved.

I hope so too. The people who are most likely to use this should be looking for trends to help mitigate risk, not for a specific hour's weather 25 days out. FWIW it did predict the East Coast heat wave in March fairly consistently.

[donsutherland1]

I hope so too. The people who are most likely to use this should be looking for trends to help mitigate risk, not for a specific hour's weather 25 days out. FWIW it did predict the East Coast heat wave in March fairly consistently.

I agree with you about using it for trends. It's encouraging to hear that the March heat wave was forecast consistently. That should give some degree of confidence going forward with the forecasts.

[RodneyS]

I hope so too. The people who are most likely to use this should be looking for trends to help mitigate risk, not for a specific hour's weather 25 days out. FWIW it did predict the East Coast heat wave in March fairly consistently.

Thanks, Jesse. Has AccuWeather ever done a study of accuracy by number of forecasting days out for different locations? For example: 5-day out forecast for April in New York City, average temperature deviation from actual, x degrees; 10-day out forecast for April in New York City, average temperature deviation from actual, y degrees; 15-day out forecast for April in New York City, average temperature deviation from actual, z degrees. If so, can you share in a general way the findings?

[Jesse Ferrell - AccuWx]

Thanks, Jesse. Has AccuWeather ever done a study of accuracy by number of forecasting days out for different locations? For example: 5-day out forecast for April in New York City, average temperature deviation from actual, x degrees; 10-day out forecast for April in New York City, average temperature deviation from actual, y degrees; 15-day out forecast for April in New York City, average temperature deviation from actual, z degrees. If so, can you share in a general way the findings?

We do internal verification to help our meteorologists produce better forecasts but, to my knowledge, we haven't done anything that complex on a by-city basis.

[OKpowdah]

What resources are used in the development of the 15-25 day range?

[gil888]

We do internal verification to help our meteorologists produce better forecasts but, to my knowledge, we haven't done anything that complex on a by-city basis.

I would be very interested in seeing the verification as compared to climatology. The east and central US heat wave was very predictable a couple of weeks in advance (perhaps not the extreme magnitude) for whatever reason. It is hard for me to believe there is significant skill added from Day 16-25 from climatology... but I'm willing to listen

[MGorse]

It's there, click on "Month"

We'll certainly be tracking the accuracy internally as we do with our 15-day forecast.

Do the forecasters actually look at the forecast for all 15 days and adjust as needed? Even out to 25 days now? Blended model approach inserted into the database this far out in time?

[OKpowdah]

Do the forecasters actually look at the forecast for all 15 days and adjust as needed? Even out to 25 days now? Blended model approach inserted into the database this far out in time?

What models do they have at their disposal for the 16-25 day range?

[MGorse]

What models do they have at their disposal for the 16-25 day range?

No idea, that is why I asked. I am not sure if they are running something internally that then gets posted to their database for those time ranges. I cannot imagine the forecasters are actually looking at the database out to 15 days let alone to 25 days for everywhere.

[Jesse Ferrell - AccuWx]

What models do they have at their disposal for the 16-25 day range?

Unfortunately, the big cheese says I can't tell (I asked). I'm sure someone here is familiar with what forecast data would be out there in that range. We take that data & historical data and run it through algorithms to improve the accuracy and resolution (in addition to any corrections that would already be in the database for our forecast points. Our meteorologists can edit the data,

It's a new frontier, for sure, but so was the 5-day when we introduced it in the 1960's. And what better place to get feedback on accuracy than the whole world?

[Cheeznado]

No offense to the Accu-WX employees who post here, but in my opinion this is 100% bunk. I have been in this game for over 30 years, and a forecast this far out is a total fantasy, plain and simple. It is just another way to "differentiate us us from the competition" or some other such corporate blather which is code for "ways we can rip folks off with a worthless product". I do a forecast to seven days for my job, and almost always days 6 and 7 tun out to be dead wrong. Specific temperatures and weather to 25 days? It is to laugh, but is also sad in a way.

[Ellinwood]

Deterministic forecasts FTL. You might be able to get away with doing temperatures, but precipitation to the nearest hundredth?

Probability forecasting is the way to go in that range, and even day-to-day probabilities are sketchy. I personally don't find any value in these forecasts in their current format.

[Cheeznado]

Deterministic forecasts FTL. You might be able to get away with doing temperatures, but precipitation to the nearest hundredth?

Probability forecasting is the way to go in that range, and even day-to-day probabilities are sketchy. I personally don't find any value in these forecasts in their current format.

OMG they have this? (April 26 in NYC)

Variable clouds with showers and thunderstorms

• Max UV Index: (low)
  
• Thunderstorm Probability: 55%
  
• Amount of Precipitation: 0.02 in
  
• Amount of Rain: 0.02 in
  
• Amount of Snow: 0.0 in
  
• Amount of Ice: 0.00 in
  
• Hours of Precipitation: 1 hrs
  
• Hours of Rain: 1 hrs
  

Thunderstorms with .02".... Wow, I guess the heart of the storm myst have just missed them- what accuracy! No, what crap. No other way to sugarcoat it. I think our profession has jumped the shark.

[MGorse]

Unfortunately, the big cheese says I can't tell (I asked). I'm sure someone here is familiar with what forecast data would be out there in that range. We take that data & historical data and run it through algorithms to improve the accuracy and resolution (in addition to any corrections that would already be in the database for our forecast points. Our meteorologists can edit the data,

It's a new frontier, for sure, but so was the 5-day when we introduced it in the 1960's. And what better place to get feedback on accuracy than the whole world?

Thanks Jesse for the added info. However, I am not sold on specific forecasts going this far out in time. Sometimes we cannot even get day 1 correct.

[Jesse Ferrell - AccuWx]

No offense to the Accu-WX employees who post here, but in my opinion this is 100% bunk. I have been in this game for over 30 years, and a forecast this far out is a total fantasy, plain and simple. It is just another way to "differentiate us us from the competition" or some other such corporate blather which is code for "ways we can rip folks off with a worthless product". I do a forecast to seven days for my job, and almost always days 6 and 7 tun out to be dead wrong. Specific temperatures and weather to 25 days? It is to laugh, but is also sad in a way.

No way, I've seen it and this is all science, not voodoo. The data is out there (it is NOT free however, which is why take this seriously as a business venture), we're just better use of it.

In talking to some other mets who have posted here today, I think the disagreement is an approach for the masses versus one for researchers. The public doesn't want "a 20% chance of temperatures between 49 and 51" -- they want to hear 50. It may be a sad fact for us scientists, but it's true. I'll retell a story that I posted on a blog earlier:

We had a large client here a few years ago, that we had presented with an hour-long demo of attractive graphs of the probability of their forecast values being reached (a range of temperatures that dictated whether they could do their work or not). We were all very proud of it. They said, "Well this is all very impressive, but all we need is a Yes/No answer for each time period. Green or red, that's it." So that's what we gave them.

[Itunis]

I'm only a freshman in meteo, but my professor had us do a semester project on the verification of weather forecasts for each main provider (NWS, TWC, AccuWx). Each group did a different city, and on average the forecast past 7 days was about as accurate as climo. It would be nice to be able to have an idea of the weather 25 days in advance, but it's just not feasible with our current technology. The output of the models is only as good as the data fed to them, and without more extensive and complete observations used to initialize the models you can't expect them to pump out a semi-decent forecast for 3 weeks out.

Oh, and if anyone cares we found that the providers were largely similar in forecast verification out to 7 days.

[MGorse]

No way, I've seen it and this is all science, not voodoo. The data is out there (it is NOT free however, which is why take this seriously as a business venture), we're just better use of it.

In talking to some other mets who have posted here today, I think the disagreement is an approach for the masses versus one for researchers. The public doesn't want "a 20% chance of temperatures between 49 and 51" -- they want to hear 50. It may be a sad fact for us scientists, but it's true. I'll retell a story that I posted on a blog earlier:

We had a large client here a few years ago, that we had presented with an hour-long demo of attractive graphs of the probability of their forecast values being reached (a range of temperatures that dictated whether they could do their work or not). We were all very proud of it. They said, "Well this is all very impressive, but all we need is a Yes/No answer for each time period. Green or red, that's it." So that's what we gave them.

Seems to be many non-believers so far doing this out to 25 days based on the comments below your announcement.

http://www.accuweath...-forecast/63449

[SerialDerecho]

OMG they have this? (April 26 in NYC)

Variable clouds with showers and thunderstorms

• Max UV Index: (low)
  
• Thunderstorm Probability: 55%
  
• Amount of Precipitation: 0.02 in
  
• Amount of Rain: 0.02 in
  
• Amount of Snow: 0.0 in
  
• Amount of Ice: 0.00 in
  
• Hours of Precipitation: 1 hrs
  
• Hours of Rain: 1 hrs
  

Thunderstorms with .02".... Wow, I guess the heart of the storm myst have just missed them- what accuracy! No, what crap. No other way to sugarcoat it. I think our profession has jumped the shark.

LOL. A 55% chance of a thunderstorm 22 days out? Can they predict the lottery numbers too?

[Jesse Ferrell - AccuWx]

Seems to be many non-believers so far doing this out to 25 days based on the comments below your announcement.

http://www.accuweath...-forecast/63449

Ahh but they're all researchers and meteorologists... not our target audience. We don't mean to be disrespectful to that group, but what they're suggesting -- holding off until it's 100% accurate or giving a statistical analysis with the forecast -- is not what the public wants. It's a bittersweet realization for me.

[mappy]

The majority of the public do not know the difference between a watch and a warning, so I guess having them believe a 25 day forecast that tells the UV index, high, low, precip to the hundredth, wind and if there will be a thunderstorm or not.... Is perfect for accuweather.

[H2O]

Ahh but they're all researchers and meteorologists... not our target audience. We don't mean to be disrespectful to that group, but what they're suggesting -- holding off until it's 100% accurate or giving a statistical analysis with the forecast -- is not what the public wants. It's a bittersweet realization for me.

But by doing that you have decided as an entity to entertain and not educate because most people will believe these 25 day forecasts to be 100% accurate and therefore blame all weather professionals when things do not verify.

[MGorse]

Ahh but they're all researchers and meteorologists... not our target audience. We don't mean to be disrespectful to that group, but what they're suggesting -- holding off until it's 100% accurate or giving a statistical analysis with the forecast -- is not what the public wants. It's a bittersweet realization for me.

They are all researchers and meteorologists posting those comments? Did you folks conduct a survey to see what the public wanted? Was going out 25 days based on public comment? Giving specific forecasts like this out to 25 days for the public to see gives the impressive that there is a lot of skill that far out, which we know is not the case.

[RodneyS]

Oh, and if anyone cares we found that the providers were largely similar in forecast verification out to 7 days.

Is this study publicly available?

[Cheeznado]

The old "the public doesn't understand the forecast anyway" argument. The bottom line here folks- this is a money making decision pure and simple. You do not really believe for a minute that any Met worth his or her salt, including Jesse who is bravely defending this product here, really deep down thinks this is worth a hill of beans. However, in the corporate world, if you can sell something worthless why not do it - integrity be damned. The venerable P.T. Barnum principle in effect.

[Ed Lizard]

Is this the CFS or the Euro weeklies (or some combined version) printed out verbatim? I got the distinct impression the 15 day AccuWx forecast was just the most recent GFS.

People in energy trading or agriculture can pay private met services for trends and outlooks, but 55% chance of thunderstorms in NYC in 4 weeks. Doesn't pass the smell test.

I doubt AccuWx private energy/agriculture/shipping clients expect that level of implied accuracy. There is a concept in engineering, 'significant figures', where the answer to an engineering problem can be no more accurate than the least accurate factor that goes into the solution. 55% chance of thunderstorms implies a level of accuracy that I doubt exists.

[Mallow]

Is this the CFS or the Euro weeklies (or some combined version) printed out verbatim? I got the distinct impression the 15 day AccuWx forecast was just the most recent GFS.

People in energy trading or agriculture can pay private met services for trends and outlooks, but 55% chance of thunderstorms in NYC in 4 weeks. Doesn't pass the smell test.

I doubt AccuWx private energy/agriculture/shipping clients expect that level of implied accuracy. There is a concept in engineering, 'significant figures', where the answer to an engineering problem can be no more accurate than the least accurate factor that goes into the solution. 55% chance of thunderstorms implies a level of accuracy that I doubt exists.

They claim it's forecasts (probably Euro weeklies, etc) run through "complex algorithms". In other words, they're doing a pseudo-MOS.