May Forecast Contest, Temperatures

[RodneyS]

DCA: 0.0

NYC: 0.0

BOS: 0.0

[bkviking]

DCA: 0.5

NYC: -1.2

BOS: -1.9

[skierinvermont]

BOS: -.2

NYC: +.7

DCA: +1.2

Also I have a question for Mallow. In the rules you say that:

One advantage to this scoring system is that if you divide your total by 300, and by the number of months in which you've participated, your average is actually a meaningful value. The average that you'd get that way is exactly equal to the average number of standard deviations you have been from the actual departures, overall.

Well let's say I had z-scores in a hypothetical month of 0,1,1. This would give me scores of 100, 32, 32 and my total would be 164. If I then divide by 300 and the number of months (1) I get .547, which is not the "average number of standard deviations you have been from the actual departures" ([0+1+1]/3=.667) . What am I missing here?

Did you mean that.547 is the number of standard deviations of error for all my guesses combined? I don't think it is but maybe I'm wrong. I don't see much meaning in this .547 number because after you convert z-scores to percentiles they are not additive.

[hudsonvalley21]

DCA: + 2.3

NYC: +2.1

BOS: +0.2

Total + 4.6

[Mallow]

Also I have a question for Mallow. In the rules you say that:

One advantage to this scoring system is that if you divide your total by 300, and by the number of months in which you've participated, your average is actually a meaningful value. The average that you'd get that way is exactly equal to the average number of standard deviations you have been from the actual departures, overall.

Well let's say I had z-scores in a hypothetical month of 0,1,1. This would give me scores of 100, 32, 32 and my total would be 164. If I then divide by 300 and the number of months (1) I get .547, which is not the "average number of standard deviations you have been from the actual departures" ([0+1+1]/3=.667) . What am I missing here?

Did you mean that.547 is the number of standard deviations of error for all my guesses combined? I don't think it is but maybe I'm wrong. I don't see much meaning in this .547 number because after you convert z-scores to percentiles they are not additive.

Nope, you're absolutely correct.

I was completely forgetting the step where I convert the z-scores to percentiles when I wrote that, so the number is more like an "average percentile". That value is not as useful as an "average number of standard deviations", and as you say, I don't even know if it's mathematically meaningful to find an average of percentiles, but I think there is at least some intuitive value there.

[skierinvermont]

Nope, you're absolutely correct.

I was completely forgetting the step where I convert the z-scores to percentiles when I wrote that, so the number is more like an "average percentile". That value is not as useful as an "average number of standard deviations", and as you say, I don't even know if it's mathematically meaningful to find an average of percentiles, but I think there is at least some intuitive value there.

Oh ok.. cool. I was doubting myself because I know you're probably more familiar with this stuff than I am. I had a great stats professor, it was a great class, but there is so much amazing math I have forgotten.

Wouldn't it be 'better' to add up our z-scores? The 'problem' with using percentiles is that someone who misses by only .2 SD gets way way more points than somebody who's off by .8 SD, but that person who's off by .8SD doesn't get that many more points than somebody who's off by a whopping 3+SD. Seems kind of skewed to me but I guess it heavily rewards those who guess the closest, sort of like the bonus system last year. The main advantage to using z-scores, as opposed to simply raw error, is that months with larger standard deviations get weighted equally as months with smaller standard deviations (otherwise the large SD months would outweigh the small SD months in determining people's scores). Obviously you're not going to change it now... but perhaps something to consider for next years contest, if you're still on the job.

[Mallow]

Oh ok.. cool. I was doubting myself because I know you're probably more familiar with this stuff than I am. I had a great stats professor, it was a great class, but there is so much amazing math I have forgotten.

Wouldn't it be 'better' to add up our z-scores? The 'problem' with using percentiles is that someone who misses by only .2 SD gets way way more points than somebody who's off by .8 SD, but that person who's off by .8SD doesn't get that many more points than somebody who's off by a whopping 3+SD. Seems kind of skewed to me but I guess it heavily rewards those who guess the closest, sort of like the bonus system last year.

The percentile method is sort of analogous to a "grade" on a true/false exam. If you're just guessing randomly (zero skill), you'll average 50%. But it definitely makes more sense to convert to a percentile AFTER averaging the z-scores (if you're looking for a good "mean percentile"). The main reason I'm using the percentiles to score the contest is because it's an easy way to get a nice round number between 1 and 100, with higher scores being "better". z-scores are less intuitive to work with from a "points" perspective.

The main advantage to using z-scores, as opposed to simply raw error, is that months with larger standard deviations get weighted equally as months with smaller standard deviations (otherwise the large SD months would outweigh the small SD months in determining people's scores). Obviously you're not going to change it now... but perhaps something to consider for next years contest, if you're still on the job.

Exactly, and that's why I incorporated them into the contest.

[skierinvermont]

The percentile method is sort of analogous to a "grade" on a true/false exam. If you're just guessing randomly (zero skill), you'll average 50%. But it definitely makes more sense to convert to a percentile AFTER averaging the z-scores (if you're looking for a good "mean percentile"). The main reason I'm using the percentiles to score the contest is because it's an easy way to get a nice round number between 1 and 100, with higher scores being "better". z-scores are less intuitive to work with from a "points" perspective.

Yeah I see what you're saying and it's probably debatable which method (z-scores or percentiles) is more 'fair' and more conforms to our expectations of who did better or worse. For example take two people, one has z-scores of 0,1,1, and the other has z-scores of 0,0,2, according to the z-score method they did equally, and just looking at their raw errors their cumulative error was about equal (provided the standard deviations are similar for all three events), but the 0,0,2 guy would have much more points in a percentile system even though his cumulative error was the same as the other guy's. That just seems kind of skewed to me. But I see what you are saying about getting it so that higher scores are better and there certainly is some intuitive value in an 'average percentile.' That's basically what one's combined SAT scores are. It doesn't make much difference and your system certainly works just fine so I will be quiet now

[Mallow]

May forecasts:

Pretty decent spread this month.

[Roger Smith]

After 10 days:

DCA -0.9

NYC +1.2

BOS +1.4

[Roger Smith]

Thru 16 days ...

DCA -0.1

NYC +0.4

BOS -0.8

cold today in Boston, not as large an anomaly for the other two.

[Inudaw]

Warm forecast not looking too good right now.

[nzucker]

Warm forecast not looking too good right now.

The last 10 days of the month look well above average, so I wouldn't say that. 0z Euro had NYC's 850s near 16C from Day 6-10, with one day approaching 20C...those are large anomalies for late May, especially if it's very humid and we don't cool down at night, the one factor that usually prevents huge warm anomalies from accumulating in spring heat waves with the generally drier air.

[Roger Smith]

After 19 days ...

DCA -0.1

NYC +0.1

BOS -1.7

let the warming begin.

[skierinvermont]

After 22 days:

BOS: -2.3

NYC: -.4

DCA: +.1

Should be able to get a good idea of the end of the month by now. Using NWS forecasts through the 29th:

BOS: -1.1

NYC: +.8

DCA: +1.9

[nzucker]

After 22 days:

BOS: -2.3

NYC: -.4

DCA: +.1

Should be able to get a good idea of the end of the month by now. Using NWS forecasts through the 29th:

BOS: -1.1

NYC: +.8

DCA: +1.9

Boston has taken quite a hit with the cut-off low...the last 11 days have had negative departures, and they're about to rack up another one today....they're sitting at a crisp 57F with fog, normal temps are 69/52 there at this time of year. NYC is also going to drop a bit today if the sun doesn't make it...only 63F in Central Park. NWS was too high today, which isn't surprising given these days with east winds and clouds tend to be more of a disaster than expected in the springtime.

Looks as if Boston finishes the month slightly below normal, NYC around normal, and DCA slightly above normal. Models have really muted the torch now.

[skierinvermont]

Looks as if Boston finishes the month slightly below normal, NYC around normal, and DCA slightly above normal. Models have really muted the torch now.

So you're saying NWS is overdoing the torch? Because otherwise BOS finishes slightly below, NYC above, and DCA considerably above.

BTW .. NWS only has the warmest days at +6 or so, which isn't exactly a huge torch. But it affects it more than you think and I doubt you've actually done the math like I did.

[nzucker]

So you're saying NWS is overdoing the torch? Because otherwise BOS finishes slightly below, NYC above, and DCA considerably above.

Well you're saying NYC would finish +0.8F if NWS is right (assuming the 30th/31st don't sway it a lot). That's "near normal" to me. No one could ever discern a month being less than 1F above the long-term average.

I could see Wednesday and Friday being a bit cooler than expected on the NWS forecast, but they look OK for the most part. Today was a massive fail as they had 73F here and we've struggled to make 60 degrees. They were also a bit warm yesterday with the projected high at 64F and the actual high verifying at a crisp 59F. But that's a pretty small point to complain about. It's spring so I expect days with E/SE winds and cloud cover to underperform on temperatures.

BTW Andrew, looks like we'll both do pretty well with both the gradient and the absolute numbers. I just need NYC to avoid the torch one day and take a big hit today.

[Mr Torchey]

The last 10 days of the month look well above average, so I wouldn't say that. 0z Euro had NYC's 850s near 16C from Day 6-10, with one day approaching 20C...those are large anomalies for late May, especially if it's very humid and we don't cool down at night, the one factor that usually prevents huge warm anomalies from accumulating in spring heat waves with the generally drier air.

rut roh zuckie:)

[skierinvermont]

Well you're saying NYC would finish +0.8F if NWS is right (assuming the 30th/31st don't sway it a lot). That's "near normal" to me. No one could ever discern a month being less than 1F above the long-term average.

I could see Wednesday and Friday being a bit cooler than expected on the NWS forecast, but they look OK for the most part. Today was a massive fail as they had 73F here and we've struggled to make 60 degrees. They were also a bit warm yesterday with the projected high at 64F and the actual high verifying at a crisp 59F. But that's a pretty small point to complain about. It's spring so I expect days with E/SE winds and cloud cover to underperform on temperatures.

BTW Andrew, looks like we'll both do pretty well with both the gradient and the absolute numbers. I just need NYC to avoid the torch one day and take a big hit today.

I haven't really looked but sometimes NWS plays it conservative on heatwaves like this so if the hottest days end up at +8 instead of +6, and the 30th and 31st end up warm like they look on the Euro, then we could be a little too low. But I think we'll do pretty well.

[nzucker]

I haven't really looked but sometimes NWS plays it conservative on heatwaves like this so if the hottest days end up at +8 instead of +6, and the 30th and 31st end up warm like they look on the Euro, then we could be a little too low. But I think we'll do pretty well.

The 12z ECM is actually pretty cool on the 30th and 31st, not sure where you're getting this from....you have to remember that the averages continue to climb steadily so temps in the mid-upper 70s aren't going to produce positive anomalies anymore...here are those days.

May 30th 850 map:

May 31st 850 map:

We are right on the edge of the heat ridge, so we'll have to watch this. If the trough in the Canadian Maritimes is any weaker, we'll torch a lot. And for the record, I believe your slightly warmer guess on NYC is probably better. But neither of us will end up too far off...this was an unremarkable month for weather, just normal temps and cloudy like many springs.

[skierinvermont]

The 12z ECM is actually pretty cool on the 30th and 31st, not sure where you're getting this from....you have to remember that the averages continue to climb steadily so temps in the mid-upper 70s aren't going to produce positive anomalies anymore...here are those day

I made that post right as the 12z ECM was coming out...

The ECM ensembles and GFS and GFS ensembles still show it well above average, but yes the 12z op ECM really brings the front through.

[skierinvermont]

NWS really upped their temps as they often do with these big heat waves:

Through the 24th:

BOS: -1.8

NYC: -.2

using NWS hi/lo through the 31st:

BOS: +.3

NYC: +1.8

[nzucker]

NWS really upped their temps as they often do with these big heat waves:

Through the 24th:

BOS: -1.8

NYC: -.2

using NWS hi/lo through the 31st:

BOS: +.3

NYC: +1.8

Next week may be a big heat wave but this hasn't been a heat wave at all.

Last couple of days have been in the upper 70s to low 80s here, low 80s both days for those within NYC. Perfectly normal for late May, definitely not a "heat wave." This week will certainly get us back to normal but we'll need to see big heat verify next week to get the stations way above average. EC does have 20C 850s in here so it's possible.

[uncle W]

Next week may be a big heat wave but this hasn't been a heat wave at all.

Last couple of days have been in the upper 70s to low 80s here, low 80s both days for those within NYC. Perfectly normal for late May, definitely not a "heat wave." This week will certainly get us back to normal but we'll need to see big heat verify next week to get the stations way above average. EC does have 20C 850s in here so it's possible.

the minimums are forecast to continue to run ten degrees above normal...maybe we pop a 90 here and there if the clouds don't get in the way...

[skierinvermont]

Next week may be a big heat wave but this hasn't been a heat wave at all.

Last couple of days have been in the upper 70s to low 80s here, low 80s both days for those within NYC. Perfectly normal for late May, definitely not a "heat wave." This week will certainly get us back to normal but we'll need to see big heat verify next week to get the stations way above average. EC does have 20C 850s in here so it's possible.

The past two days in NYC have been +9 and +10, not "perfectly normal." It was 84 in NYC yesterday, the average is 74, not the low 80s. That's a heat wave. Nearly every day for the rest of the month runs +6-12.

[Ellinwood]

Looks like it may come down to a battle between myself, blazess and hudsonvalley for May. Blazess and I have the potential to win/lose by 0.1

[nzucker]

The past two days in NYC have been +9 and +10, not "perfectly normal." It was 84 in NYC yesterday, the average is 74, not the low 80s. That's a heat wave. Nearly every day for the rest of the month runs +6-12.

It's above average, but it's a "perfectly normal" warm spell, not an unusual one. You need to read up on the history of NYC heat waves...we've seen mid 90s at this point in the game. 80F is nothing special for Memorial Day weekend. NYC can hit 90F in early April, so hitting 80F a month and a half later isn't impressive. It was also in the 50s/low 60s Sunday and Monday, so this is a balancing act.

[Inudaw]

Looks like it may come down to a battle between myself, blazess and hudsonvalley for May. Blazess and I have the potential to win/lose by 0.1

Yea yall look good.. maybe to warm on Boston....

I'll bust badly on Boston... and New York... Could Salvage DC...

[skierinvermont]

It's above average, but it's a "perfectly normal" warm spell, not an unusual one. You need to read up on the history of NYC heat waves...we've seen mid 90s at this point in the game. 80F is nothing special for Memorial Day weekend. NYC can hit 90F in early April, so hitting 80F a month and a half later isn't impressive. It was also in the 50s/low 60s Sunday and Monday, so this is a balancing act.

Oh so we've gone from calling it "perfectly normal" to a "perfectly normal warm spell." I'm perfectly aware of how hot NYC can get this time of year so don't lecture me on "reading up." Just because it's not record breaking doesn't mean it's not a heat wave. I called it a heat wave.. you corrected me and told me it was "perfectly normal" and "definitely not a heatwave." Now you're saying it's a "perfectly normal warm spell." What a joke. 7 days of +10 is a rare event.. I don't care what you call it.