TO ALL HAS FORECASTERS, DOHs, HICs, and HSD CHIEFS,

Introducing.... HAS NOTES... After reviewing your responses to our initial inquiry concerning names for this news letter, we have decided on HAS NOTES. There were concerns that the name HASMAT would be perceived negatively rather than humorously due to its similarity to the acronym for hazardous material. Hopefully, HAS NOTES will be descriptive of what is actually contained within, which are simply notes for and from HAS forecasters.

In addition to providing useful information from HRL, we want to make the HAS NOTES a useful place for HAS forecasters to share information about issues affecting their HAS function with HRL and other RFC's. It is clear that there is a wide variation in the way the HAS function is run between the various RFC's. Hopefully we can all use HAS NOTES as a forum to discuss and organize our efforts so that we can all be more effective both from a national perspective as well at individual RFCs. If you have some material that you would like to include in the next issue of HAS NOTES please E-mail your write up to me by June 1st and I will include it under: NOTES FROM XXRFC. To keep things simple, just include the information in the body of your E-mail message, or send as an attachment (ASCII or wordperfect), which ever is easier for you. My CC-mail address is:

jay.breidenbach@noaa.gov

In this issue, after reading NOTES FROM HRL, be sure to check out NOTES FROM OHRFC where Jim Noel and Mark Fenbers discuss the use of the "Precipitation Potential Factor". Under NOTES FROM WGRFC you will find an interesting summary of issues affecting the HAS unit at WGRFC and a summary of their work on QPF and MAPX verification.

NOTES FROM HRL

We have received quite a bit of feed back on the Stage III rainfall estimates indicating RFC wide underestimation when compared with raingage estimates. We are acutely aware of the problem from several Stage III case studies as well as other validations performed at HRL. There are several reasons for this systematic underestimation. The primary cause for this underestimation is that the Stage I estimates are systematically low due to: Use of inappropriate Z-R relationships, uncalibrated radars, severe range degradation in statiform rainfall due to sloping reflectivity gradient in the vertical, incomplete beam filling and beam overshoot of precipitation, beam blockage due to terrain, and wet radome attenuation in very heavy rainfall. HRL and cooperative researchers (Professors Jim Smith at Princeton University and Witek Krajewski of the University of Iowa) are currently involved in research to find possible solutions to these problems in the Stage I algorithm. Currently, however, there is little that the RFC can do to change these biases other than to recognize when and where they exist. Since the turning-off of biscan maximization, the severity of underestimation has gotten worse: Prof. Jim Smith is expected to provide guidances on this matter in the next two months (possibly a new Z-R relationship).

Another part of the underestimation problem can be attributed to less than optimal bias adjustment in Stage II, and can partially be addressed by optimizing the current stage II parameters. The new stage II bias adjustment procedure and Stage III mosaicking algorithm due to be released by HRL later this month (if beta-testing at OHRFC goes successfully) should significantly improve Stage III estimates except for gage-poor sites.

The Stage II algorithm is designed to handle some of under- and overestimation problems with its bias adjustment algorithm and multisensor radar/gage merging technique. However, in the case study work mentioned above, we found that the bias produced by the Kalman filter technique was not responding fast enough to the actual sample bias. The Kalman filter bias always started at 1.0 and maybe, by the end of the event, was near the actual sample bias. The cumulative result was that the Stage II estimate, although not as bad as stage I was still an underestimate. The case was re-run using new adaptable parameter settings for the bias adjustment algorithm so that Kalman filter would respond to the sample bias much quicker. The new results were much better. (i.e. less bias, less under-estimation). The parameters that were changed are:

Parameter Default Test Value

S2_correlation 0.1 1.0

Stage_eqn_variance 1.0 0.1

Caution: The net effect of changing these parameters is that the Kalman Filter will produce a bias that is almost identical to the sample bias each hour. This may be unrealistic for small gage networks and could cause instabilities in the algorithm leading to unrealistically large stage II biases. For radars with a large number of gages > 100, the test settings will work fine, however, for smaller gage networks, conservative values closer to the default will insure stability in the resulting Kalman filter bias.

HRL recommendation: Wait for the new stage II bias adjustment algorithm due out soon which does a better job at mass-balancing. A single adaptable parameter, alpha, will control how much weight is placed on the most recent gage radar pairs in computing the stage II bias (details will follow in the release note).

There are two other Stage II parameters that have an effect on the bias in the vicinity of rain gage observations (ml_lag0_ind_corr and ml_lag0_cond_corr). The ml_lag0_ind_corr parameter indicates on a scale of 0 to 1 how much confidence you have in the radar verses raingage being able to define where it is raining and where it is not. A setting of 1 means that you have complete confidence in the radar's ability to define the spatial coverage of rainfall, while a setting of 0 means that you would place complete faith in the rain gage field representation of where it is raining. For convective situations, the radar is very good atdefining where it is raining, therefor a setting near 1 is appropriate. However, for more uniform, stratiform type situations, the gage field gains skill, and hence a lower setting for this parameter may be more appropriate.

Given that Stage II has determined that it is raining at certain HRAP grid points, the ml_lag0_cond_corr parameter defines how much confidence on a scale from 0 to 1 you have in the radars ability to determine the rainfall accumulation at those grid points. A setting of 1.0 means that the radar estimate is used at all grid points except the ones where a gage observation falls within the grid point. As ml_lag0_cond_corr is decreased the influence of the gage value on surrounding grid points increases. If ml_lag0_cond_corr is set all the way down to 0, the radar will have no influence on the value of any HRAP grid that falls within the influence range of any gage, therefore, the Stage II analysis will look like the gage only analysis.

In a case study with lots of raingages, for a stratiform rainfall event, the following test values improved the rainfall estimate and reduced the bias and rms error when compared with independent gage observations.

parameter default test value

ml_lag0_ind_corr 0.85 0.5

ml_lag0_cond_corr 0.80 0.5

Caution: For radars that do not have very many raingages, lowering these parameters will likely degrade your stage II rainfall estimate. Smaller values will also cause degraded estimates if the character of the rainfall event is more convective than statiform.

HRL recommendation: For radars with lots of gages (> 100), lowering the value of ml_lag0_ind_corr and ml_lag0_cond_corr below the default, should improve rainfall estimates during the winter months when rainfall tends to be more stratiform in nature. For radars with few gages, or during the summer convective season, the default values should work best.

NOTES FROM OHRFC HAS FORECASTERS

The OHRFC has been using a derived parameter call Precipitation Potential

since 1996.

This parameter multiplies the 1000-700 mb mean relative humidity by the

Precipitable Water (PW). Units are in inches of PW. This combination was

developed by Rod Schofield of NESDIS for use with satellite estimations of

rainfall and he called it precipitation efficiency. Rod uses satellite-

derived PW estimates with measured mean RH to identify short-term flooding

threats. He uses this parameter to see how efficient the environment is

to correct satellite estimations of rainfall; hence precipitation efficiency.

Mark Fenbers of OHRFC took the concept a step further to use in conjunction

with model forecasts of PWs/Mean 1000-700 RHs. This parameter is called

Precipitation Potential as it uses model objectively analyzed schemes and

model forecasts of PWs and Mean RHs. This parameters goal is not to see

how efficient the environment is in order to make corrections to satellite

derived rainfall estimates but to try to forecast rainfall, QPF. Since this

technique is a forecast parameter, uncertainity from the model creeps in.

In the case of use with realtime data from Rod's technique, there really

is no uncertainity that rainfall is occurring. However, with model data,

since so much revolves around the dynamics and thermodynamics of the

atmosphere in order to generate rainfall and there is no guarantee even with

many parameters coming together, we are talking about a rainfall potential

and not a certainity. Hence the name Precip Potential.

A deep moist layer is much more efficient at raining out the PW in the

atmosphere than a dry layer. It can also be inferred that deep moisture is

indicative of some low-level convergence and some rising motions.

Precipitation Potential assumes that the necessary ingredients are

there for rainfall to be produced and does not handle isolated maximum

amounts.

Our goals as HAS forecasters is forecast where it will rain, how much and

get it in the right location. Our observation is that Precipitation

Potential is useful at this. Our observation is that the maximum rainfall is

often located near maximum Precip Potential values. In addition, there is

a remarkable relationship between precipitation potential and actual

rainfall. When these values exceed 1", moderate or heavy rainfall can occur.

When these values exceed 1.5", the threat for heavy rainfall is greatly

increased.

Another useful tool of Precip Potential is locating the leading edge of

precipitation with the leading edge of an organized rainfall event. In the

cool season, this is locate near the 0.4" to 0.5" line. In the warm season,

it is located near the 0.6" to 0.8" line for the Ohio Valley. This will

likely vary a little depending on your geographic location.

Remember, we are still pretty inexperienced with this new Precipitation

Potential derived parameter. We do plan on doing more research on this

parameter in the near future.

The following the FDF for GARP to implement at your office.

! GARP Field Description File (FDF) Template

!

! Field: Precipitation Potential (inches).

!

! P.Bruehl/NWS

! *** Required Parameters ***

type = scalar

layer = n

function =

mul(quo(pwtr%SGMA@0:10000;2540),quo(add(add(relh,relh@850),relh@700),3))

description = Precipitation Potential was implemented at OHRFC for

model forecasts and developed by Mark Fenbers and

adapted from Rod Scofield's Precipitation Efficiency.

This parameter takes PWs and multiples it by the

1000-700 mb mean RH to see how efficient the

atmosphere is to raining out the all the pw.

(PW X MEAN 1000-700RH).

= Sigma levels 0 through 1

! *** End of Required Parameters ***

!

! *** Significant Parameters ***

label =

vcoord =

level1 = 1000

level2 =

contour = y

fill = y

hilo = n

! Pressure coordinates

pres =

pres_cint = 0.2

pres_cmin = 0.4

pres_cmax = 2.6

pres_fint = 0.2

pres_fmin = 0.4

pres_fmax = 2.6

! Sigma coordinates

sigma =

sigma_cint = 0.2

sigma_cmin = 0.4

sigma_cmax = 2.6

sigma_fint = 0.2

sigma_fmin = 0.4

sigma_fmax = 2.6

! Theta coordinates

theta =

theta_cint =

theta_cmin =

theta_cmax =

theta_fint =

theta_fmin =

theta_fmax =

! User defined coordinates

anyvcoord =

anyvcoord_cint =

anyvcoord_cmin =

anyvcoord_cmax =

anyvcoord_fint =

anyvcoord_fmin =

anyvcoord_fmax =

! LINE

line_color = 9

line_type = 1

line_width = 1

line_label_frequency = 1

! SCALE

scale = 0

! HILO

colorh = 1

colorl = 1

symbolh = H#

symboll = L#

rangeh =

rangel =

radius =

counth = 30

countl = 30

interp = y

! HLSYM

sizes = 2

sizev = 1.5

position =

fonts = 21

fontv =

widths =

widthv =

hwflgs = hw

hwflgv =

! CLRBAR

colorbar = 1

colorbar_orientation =

colorbar_anchor =

colorbar_x =

colorbar_y =

colorbar_length =

colorbar_width =

colorbar_frequency =

! CONTUR

subbox =

smoothing = y

! TEXT

text_size =

text_font =

text_width =

text_hw_flag =

! FLINE

fline = 0;22;21;20;19;18;17;16;15;14;13;13;13;13;13;13;13

! SKIP

skip_contour =

skip_plot_x =

skip_plot_y =

! *** End of Parameters ***

An initial attempt is being made to forecast rainfall amounts > 0.5" at the

OHRFC by Jim Noel with Precip Potential with the following technique:

A. The environment is not efficient at producing precipitation below 0.5"

Precip Potential. Therefore, subtract 0.5" from Precip Potential for a

6-hour rainfall total.

B. If it is a stratiform event, divide by 2.

C. If it is a convective event, multiple by 2.

This technique assumes the possibility for training and lingering of rainfall

as long as the model forecasts it. If a combination of stratiform and

convective rainfall is anticipated, using Precip Potential minus the 0.5" is

your best bet.

Please call us if you find some other factors or adjustments that

work better or if you need help getting it into GARP/Ntrans/PC-GRIDDS.

Jim Noel/Mark Fenbers

OHRFC

NOTES FROM THE WGRFC HAS FORECASTERS

The HAS Forecasters have been quite busy of late at the West Gulf River Forecast

Center in Fort Worth, Texas. One reason we've been so busy is that we have been

running with only two HAS since mid-January, 1997. Many of you know Patricia

Brown, WGRFC Senior HAS Forecaster from October, 1993 to January, 1997. Pat got

a position as a meteorologist (forecaster) at the WSFO in Jackson, Mississippi.

For Pat, this was a homecoming for her, as she went to college in

Jackson, and still has many friends and relatives there. We wish her the best in

her new endeavors. Meanwhile, it is not known how long it will be until a new

Senior HAS is named at WGRFC.

Since Pat left, the big Texas rains began, which left the other two HAS, Tony

Hall and Greg Story, to "pick up" the slack. February, 1997 saw record rainfall

amounts in several North Texas cities, including the Dallas/Fort Worth

metroplex. Minor to moderate flooding became fairly widespread. The rainfall

slowed up a bit in March, but by late March into early April the big rainfall

returned, with Victoria Texas receiving over 10 inches of rain over a four day

period. Then on April 4 and 5 portions of northwest Louisiana received 6 to 10

inches of rain. So again, minor to moderate flooding resulted.

Aside from working shifts, Tony and Greg have been working on many problems we

deal with at WGRFC. Since we get such large rainfall amounts, Tony has been

working on different programs dealing with QPF. Also, a problem we deal with

almost on a daily basis is the underestimation of rainfall from the WSR-88Ds in

the WGRFC area of responsibility, especially those radars close to the Gulf of

Mexico. This also involved a bias found in the Stage III Precipitation

Processing System which we use which seemed to magnify the underestimation of

the radars. Greg has been examining these problems. Lastly, we are beginning a

verification program for QPF. We will be comparing the future MAP output from

HAS_QPF with the observed MAPs. In addition, we hope to compare forecast river

stages which use QPF as opposed to not using QPF. Greg is the verification focal

point for these projects.

Tony Hall has done a lot of work with QPF. He started with the development of

"heavy rain" macros using PCGRIDDS. This also includes the use of the Deep

Convection Index, which was developed by a forecaster at the WSFO in Topeka,

Kansas. These macros point us to the potential "hot spots" where rainfall may be

heavier than the model QPF may suggest. These macros are used operationally on

an almost daily basis (depending on the expected weather).

Then Tony began an experiment using a software called Brainmaker. This software

uses neural networks to make predictions. After accumulating three years' worth

of meteorological and climatological data for the Dallas/Fort Worth metroplex,

Tony "trained" the neural networks with the idea of having Brainmaker make QPF

and POP forecasts. The results have been very surprising. They consistently are

more accurate than any of the models' QPF. To read all about Tony's work with

neural networks using the Braimaker software, you can see a paper he wrote

through the WGRFC Home Page. The URL address for our home page is:

http://www.srh.noaa.gov/wgrfc

To go directly to Tony's paper, the URL is:

http://www.srh.noaa.gov/wgrfc/brainmaker

This paper originally appeared in the April 1, 1996 edition of Southern Region

Topics as a technical attachment. The Brainmaker software is run on a daily

basis to see what the neural network prediction for QPF and POP in the

Dallas/Fort Worth area is.

Next, Tony launched an extensive study on the accuracy and the operational

utility of the AFOS 94Q graphic. If you wish to see the results of this study,

which appeared in the December 1, 1996 edition of Southern Region Topics as a

technical attachment, it is also available on the internet. Its URL is:

http://www.srh.noaa.gov/wgrfc/brainmaker/94q_study.html

Tony's latest project he is undertaking has to do with the accuracy of the QPFs

from the Eta, NGM, and the AVN models. This study is ongoing, and has yet to be

written or published.

One of the concerns that the WGRFC HAS Forecasters have had is that the WSR-88D

Precipitation Estimates have been low when compared to rain gage observations.

About the only time the WSR-88Ds in Texas would overestimate the amount of

rainfall, when compared to ground truth, is when there is the presence of frozen

precipitation, either at the surface or aloft. The problem appears to be worse

as you get further south in Texas and Louisiana near the Gulf of Mexico where

the air mass is mostly tropical in nature. Greg Story has conducted several

studies with this problem in mind.

One paper he wrote was addressed to the WFOs which are within the WGRFC area of

responsibility. It explains how the Stage II and Stage III Precipitation Process

works. With this information, it is hoped that the WSR-88D operators will know

how changes to their radar affect the precipitation algorithm and the estimates

of their radar. If you would like to see a modified version of this paper, which

appeared in the January 15, 1997 edition of Southern Region Topics,

you can find it on the internet at this URL:

http://www.srh.noaa.gov/wgrfc/hdp/hdp_paper.html

Another study Greg is conducting compares the Mean Areal Precipitation output

using the old Theissen method of calculating MAP with MAPX, which is the MAP

output from Stage III. While the old MAPs have their own unique set of problems,

it has in the past been considered the standard by WGRFC Hydrologic Forecasters.

The comparisons showed that the new MAPX values from Stage III were consistently

lower than the old MAPs. After further study, it was determined that the radars

underestimation problem, combined with the way the Stage III program arrived at

its one-hourly estimates, have combined to produce the understimation of

rainfall contained in the MAPX values. We have talked extensively with the

Hydrometeorology group at OH, and they are coming out with a new build of the

Stage III software later this spring to alleviate some of the Stage III

precipitation processing shortcomings. In short, what the new version of Stage

III will do is give the HAS Forecaster the option to chose either the highest

rainfall estimate from each HRAP grid where radar umbrellas overlap, or the

averaging method. Presently, where two radars overlap, the assumption is made

that if one radar multisensor field is underestimated, the other radar must be

overestimating by the same factor in order for Stage III to come up with a

correct estimate. We have found this to be a false assumption.

The problems with the WSR-88Ds underestimating precipitation are well known. One

of the things Greg has encouraged the radar sites along the Gulf Coast to do is

to try a different Z-R relationship. For some time, all radar sites have been

equipped to change from the "default" Z-R relationship to a "tropical" Z-R

relationship. Both the radars in Lake Charles, LA and Houston, TX have changed

to the tropical Z-R relationship, and the precipitation estimates in heavy rain

situations have been greatly improved.

Lastly, while our QPF verification studies are just beginning, the preliminary

findings show that the forecast QPF values from the WFOs are usually too high

(except in extremely heavy rainfall situations), that the QPF is usually too

widespread in areal coverage, but that the probability of detection by each of

the WFOs is high.

We look forward to hearing about what the HAS units at other RFCs are working

on, and what problems you have encountered in your daily HAS shifts.