Dear all,

I am new to stats and would be grateful if someone could kindly help
me out with what is probably a basic question.

Given daily rainfall amounts below for a single month (999 indicates
missing data).

I would like to know what the probability of wet day following a wet
day is and the prob of a dry day following a wet one is.


I think Pww=0.50, Pwd=0.22

....or is Pww=0.15.

I wish to set up a weather generator and require the transition
probabilities.

Any help much appreciated.


Rees

DATA
----------------
0
0.854
0.487
2.06
0.392
0
2.292
0
1.773
0
0.606
0.298
0
0
0
0
999
999
999
0
0
0
0
0
0
0
0
0
0

Rees wrote:

Dear all,

I am new to stats and would be grateful if someone could kindly help
me out with what is probably a basic question.

Given daily rainfall amounts below for a single month (999 indicates
missing data).

I would like to know what the probability of wet day following a wet
day is and the prob of a dry day following a wet one is.

I think Pww=0.50, Pwd=0.22

...or is Pww=0.15.

I wish to set up a weather generator and require the transition
probabilities.

Any help much appreciated.

Rees

DATA
----------------
0
0.854
0.487
2.06
0.392
0
2.292
0
1.773
0
0.606
0.298
0
0
0
0
999
999
999
0
0
0
0
0
0
0
0
0
0

I get Pww as 4/28, or about .14, and Pwd as the same. Pdd is 13/28
and Pdw is also 4/28 (ignoring the missing days). While this is a
good problem for learning something about statistics, it will give
a rather poor rainfall generator compared to reality for most
locations.

Regards,
Russell

"R. Martin" wrote in message ...
I get Pww as 4/28, or about .14, and Pwd as the same. Pdd is 13/28
and Pdw is also 4/28 (ignoring the missing days). While this is a
good problem for learning something about statistics, it will give
a rather poor rainfall generator compared to reality for most
locations.

Regards,
Russell

If you're saying the data doesn't reflect prior knowledge, then one
should NOT be ignoring the prior knowledge. And if the data is not a
good match for reality, why use it at all?

If the prior is so good and the data so bad, just use the prior and be
done with it.

Glen

(Glen) wrote in message . com...
"R. Martin" wrote in message ...
I get Pww as 4/28, or about .14, and Pwd as the same. Pdd is 13/28
and Pdw is also 4/28 (ignoring the missing days). While this is a
good problem for learning something about statistics, it will give
a rather poor rainfall generator compared to reality for most
locations.

Regards,
Russell

If you're saying the data doesn't reflect prior knowledge, then one
should NOT be ignoring the prior knowledge. And if the data is not a
good match for reality, why use it at all?

If the prior is so good and the data so bad, just use the prior and be
done with it.

Glen

I'm saying exactly what I said, that that simple *model* will not be a
good one. This has been known for a long time because such models were
abandoned a long time ago, to be replaced with more sophisticated models
that work better. Prior knowledge must be used, but there is much more
potential information available in the prior knowledge than is
incorporated in such a simple model.

Regards,
Russell

You could fit a lag 1 regression model where you predict y (todays
precipitation) from X (yesterdays precipitation). But rather than
estimating the mean in a linear model, I would estimate all the
quantiles with a flexible nonlinear approach such as cubic splines. See
Roger Koenker's web page (www.econ.uiuc.edu~roger) and a short course
example on a similar problem related to daily maximum temperatures (in
Melbourne, Australia).

Brian Cade (USGS)


Rees wrote:

Dear all,

I am new to stats and would be grateful if someone could kindly help
me out with what is probably a basic question.

Given daily rainfall amounts below for a single month (999 indicates
missing data).

I would like to know what the probability of wet day following a wet
day is and the prob of a dry day following a wet one is.

I think Pww=0.50, Pwd=0.22

...or is Pww=0.15.

I wish to set up a weather generator and require the transition
probabilities.

Any help much appreciated.

Rees

DATA
----------------
0
0.854
0.487
2.06
0.392
0
2.292
0
1.773
0
0.606
0.298
0
0
0
0
999
999
999
0
0
0
0
0
0
0
0
0
0