Stephen Stein wrote:

I won't comment about the rest of the Artist's post, but I actually would be
curious to see what TWC would say about this.

Then e-mail them. I e-mailed them regarding a couple things and as I recall
only received an automated message acknowledging it. This is their problem, not
mine. I understand how to present probabilities - though am not always good
at guessing them.

But Joseph, I wonder when you question the accuracy of hourly forecasts.
I'd expect them to be better about things like precip prob at a well-known
location like Gettysburg. But if you're talking about the hourly forecasts
given for a specific zipcode, I wouldn't expect much in the way of accuracy.
Especially for a place like the Poconos, where I'm sure there's a large
variability throughout the area.

For the case I was mentioning, the specific location was not an issue. For
that, they clearly overestimated the potential for rain - no different than
my 70 % for Thursday.

The thing I am criticizing is forecasting HOURLY probabilities which are as
great or sometimes greater than DAILY ones. That is mathematically IMPOSSIBLE.

For example, suppose they forecast a 40 % chance of rain for each hour between
12 & 7 PM. If the events were independent (i.e., showers were random, began
and ended immediately, and had that probability any hour), then the chance of
a shower during that entire period is 1 - .6^7 = .972, or 97.2 %. I.e., the
chance of not getting a shower one hour is .6 - 2 hours is .6*.6, etc. For
7 hours, the chance of not getting a shower is thus (.6*.6*.6*.6*.6*.6*.6) =
...6^7 = .028 . So the chance a shower does occur is 1 -.028 = .972 .

Showers are not independent events, nor are they instant. They linger, and
often occur in general areas. So the estimation of probabilities is to some
extent an issue of forecasting where those areas of greatest activity will
be, but it is also to some extent random as described above. So if the
hourly probabilities were 40 %, 97.2 % would be an upper bound on the
probability during the entire 7-hour period - it would actually be lower.
40 % would of course be a lower bound. If the probability of any hour is
40 %, the probability during the entire period must be at least that (some
of TWC forecasts don't obey this simple rule).

For synoptic rain or snow events, the hourly probabilities can often be
close to the daily ones because an entire storm can either hit or miss a
location. So in that case, you are essentially saying there's a 40 % chance
the storm does affect your region (and if it does so, you'll probably get
precipitation most of those hours). But that is not the case for summertime
showers. For them, each hourly probability should generally be much less
than the daily one.

So to summarize : for the above example, if there's a 40 % chance of a shower
each hour, there must be between a 40 - 97.2 % chance of at least one during
the entire 7-hour period. 80-85 % would probably be typical.

If the daily probability is 40 %, the hourly ones should generally be about
10-20 % at most - some hours as low as 5 %. Experience supports this, because
one brief shower does not ruin a whole day. If you are on the golf course,
you duck under a shelter for 20 minutes and then continue your play. 40 %
chance of a shower - it happens - and the rest of the day is fine. That's
what those low probabilities represent, and why it is misleading to place
probabilities of 30-40 % each hour if the daily one is 40 %.

I live in Acton MA, but as far as I know, the TWC forecasts are all based on
Bedford, which is about 10 miles closer to the coast. I don't know about
other weather stats, but I know that snow amounts can vary tremendously
between those locations.

I can't say much about the 10 mile zone, but something I mentioned previously
on ne.weather is that snow cannot blow or drift from the ocean to land, but
does across land. Noreasters often have strong inland winds, so it makes
sense that snow will not collect well right along the coast, and generally
advances inland (along the ground). I don't know to what extent this is
responsible for the observed differences of amounts.

To me, these forecasts are like extra decimal digits beyond the limits of
significance in a physics problem answer, when you really don't have enough
data to make a prediction beyond regional daily conditions. In that way, I
think that publishing hourly predictions is misleading. Perhaps TWC should
limit itself to what it can reasonably predict.

There are local effects which are predictable, but I don't think they are
included in the model(s) they use. Nor do I always get them right here in
Mount Pocono. We had a dark cloud overhead which lingered almost all day
yesterday - we were the last spot to clear, and probably the coldest spot
in the state. I had a maximum of 72°, which was a ° lower than MPO on the
plateau (we are usually a bit warmer this time of year, though late in the
year when vegetation is more abundant, we are often a bit cooler during day).
Yet during a warm, sunny morning with warm air aloft, we here near the top
of the slope facing ESE in Mount Pocono can be the warmest spot in the state
around 10-11 AM. In that situation, we are often warmer than PHL & ABE.
We are east, so heat a little sooner than western locations of the state
also. The hourly forecast from TWC and the NWS graphics don't show this
at all - but you certainly feel it.


--
This article was auto-posted by the ne.weather.moderated Weatherbot
program. The author is solely responsible for its content.

ne.weather.moderated FAQ/Charter: http://www.panix.com/~newm/faq.txt
ne.weather.moderated moderators e-mail:
(Please put "wx" or "weather" in the subject line to avoid the spam block.)

On 15 May 2004 03:52:25 -0400, Joseph Bartlo
wrote:


For the case I was mentioning, the specific location was not an issue. For
that, they clearly overestimated the potential for rain - no different than
my 70 % for Thursday.

The thing I am criticizing is forecasting HOURLY probabilities which are as
great or sometimes greater than DAILY ones. That is mathematically IMPOSSIBLE.

snip

The NWS gets around this in their tabular forecasts by placing a
qualitative description in the precip block. Their tabular forecasts
are experimental and aren't used in all regions. BWI is using it,
however:

http://tinyurl.com/yqsu8

Note that they differentiate between rain and thunder.


--
This article was auto-posted by the ne.weather.moderated Weatherbot
program. The author is solely responsible for its content.

ne.weather.moderated FAQ/Charter: http://www.panix.com/~newm/faq.txt
ne.weather.moderated moderators e-mail:
(Please put "wx" or "weather" in the subject line to avoid the spam block.)

On 15 May 2004 13:22:37 -0400, The Artist Formerly Known As Your
Highness wrote:

On 15 May 2004 03:52:25 -0400, Joseph Bartlo
wrote:


For the case I was mentioning, the specific location was not an issue. For
that, they clearly overestimated the potential for rain - no different than
my 70 % for Thursday.

The thing I am criticizing is forecasting HOURLY probabilities which are as
great or sometimes greater than DAILY ones. That is mathematically IMPOSSIBLE.

snip

The NWS gets around this in their tabular forecasts by placing a
qualitative description in the precip block. Their tabular forecasts
are experimental and aren't used in all regions. BWI is using it,
however:

http://tinyurl.com/yqsu8

Note that they differentiate between rain and thunder.


Whoops, it's not BWI, it's Sterling, VA's NWSFO that's issuing this
product.


--
This article was auto-posted by the ne.weather.moderated Weatherbot
program. The author is solely responsible for its content.

ne.weather.moderated FAQ/Charter: http://www.panix.com/~newm/faq.txt
ne.weather.moderated moderators e-mail:
(Please put "wx" or "weather" in the subject line to avoid the spam block.)

The Artist Formerly Known As Your Highness wrote:

The NWS gets around this in their tabular forecasts by placing a
qualitative description in the precip block. Their tabular forecasts
are experimental and aren't used in all regions. BWI is using it,
however:

http://tinyurl.com/yqsu8

Note that they differentiate between rain and thunder.

I think they'd find when verifying these forecasts that those hourly PoP's
are much too high also. Most summertime rain is showers - how many hours of
a day does that verify correctly?


--
This article was auto-posted by the ne.weather.moderated Weatherbot
program. The author is solely responsible for its content.

ne.weather.moderated FAQ/Charter: http://www.panix.com/~newm/faq.txt
ne.weather.moderated moderators e-mail:
(Please put "wx" or "weather" in the subject line to avoid the spam block.)