Dear Martin,

In looking over the new plots I have a suggested slightly revised
outline and figure set.

By the way, did I send you my powerpoint talk from a precipitation
conference last July which addresses some of these issues. I cannot
remember.

I will be in Seattle on Thursday so I can call you from there.
Let me know what time and also the number.

Regards,
Sandra

1. Introduction

General dicussion of value of rainfall mapping with radar
and also basic types of errors associated with it (ref. Austin 1987). 

Disdrometer data can help in addressing errors associated with Z-R
uncertainties (one of several errors). I.e. it does not help with
calibration, blocking, difference in R profile between height of beam
and surface etc. Note that others have used Z-R to compensate for
calibration errors (i.e. radar vs. rain gauge derived).  We prefer to
treat independent sources of error independently.  In this paper we
look only at DSD-related issues.

2. Data

A. Brief description of instruments

B. Comparison of daily data to show that the data we have is good.

I suggest we do this in terms of a table rather than a figure since
a table will be easier for other PIs to refer to.

Table Daily rainfall (indicating incomplete days for UW)

columns for: 
Rain gauge (mm), DLR (mm), UW (mm), 
checkmark if both within 10% of gauge total

I don't want to have the incomplete days from the UW disdrometer stand
out as these are operator error (i.e. that I sent students and could
not be there myself) as opposed to instrument problems and I dont want
it to be misinterpreted as an instrument problem. If we use a plot we
need to set to missing those days when we dont have full 24 hours.

Explain (briefly with reference to Smith et al. 1993) sampling errors
and why we are using 10 min accumulated data for our analysis to make
the errors smaller.

Explain other sources of error like wind and calibration and how
large we think the errors might be.

Discuss possible bias in DLR data and its origin (if it is still there).

Show that the DLR and UW data represent two different samples of the
same population with Figure ZR scatter plot.

3. Analysis

Discuss (briefly) issues associated with deriving ZRs.
Explain that we will follow Doelling et al. and others in using a
fixed Z=aR^1.5.

Show a plot of R vs. log (a) to show how scattered it is and that
there are not clear sub categories by rain rate. [we need to do this plot
I have something similar for several other data sets].

Explain that based on the R vs. log(a) scatter that there is not
one "a" for the dataset. Instead we need to determine a "best a value"
based on some criteria. 

Explain our criteria, (which we need to figure out and discuss).
The two main choices are the value which yields the best
accumulated R or the best rainrate at any given time. Joss calls these
the "hydrologist" and "tourist" applications.

Show how sensitive "a" is based on details of sample population
(i.e. 2 halves of data (even and odd samples) dont yield exactly the
same result).

Show the uncertainty in the "a" value based on a standard measurement
like standard deviation. (I am a bit uncomfortable about this since
the distributions are not Gaussian).  Put these in a table with the
best and +/- values.

I like your bar plot frequ_a_10min.gif, we can use this to show what
the distribution looks like. I think this is better than the
cumulative plots to interpret. We might do another version of this
plot, with y axis the normalized contribution to total rainfall to
cover the hydrologist application.

Show that by applying the uncertainty values the rainrate can vary 
by a large amount. Refer back to Austin's statements about
only getting rainrate from radar within a factor of 2.

Compare our "best" Z-R with the current operational ones and note that
the variation is within the uncertainty (we have to check this)
and repeat that we cannot address other types of errors.

Conclusion

Advise MAP community that use of one "standard" Z-R in their analysis
will make comparisons between PIs easier BUT does not mean it is
definitive. How "good" Z-R is is function of how representative our
data sets are of the rainfall distribution they have under study.


**********