Docstoc

qpf

Document Sample
qpf Powered By Docstoc
					    QPF ISSUES IN
         NWP
          William A. Gallus, Jr.
Dept. of Geological & Atmospheric Science
           Iowa State University
ETA
                             MM5


      OBS PREC:3/12-4/12        OBS PREC:4/12-5/12




          24HR PRECIPITATION 4/00-5/00
ETA                          MM5



OBS:3/12-4/12      OBS:4/12-5/12      OBS:5/12-6/12




         48 HR PRECIPITATION (4/00-6/00)
What is “TRUTH” for QPF verification?
       GOOD NEWS: QPF is
          improving!!
•Increased computer resources have allowed
better parameterization schemes and model
resolution
•2-day precipitation forecast today is now as
accurate as 1-day forecast in 1974
•Each resolution improvement in NCEP Eta
model improves skill scores
MRF has some skill
compared to persistence,
even out to 7-8 days:


Roads et al. 1991 (WAF)
This skill is even
more apparent for
heavy rainfall cases
 BAD NEWS: Problems abound
•Most improvement in QPF scores occurs
during cold season - little improvement in
warm season
•Flash flooding kills more people than any
other convective-related event
•QPF problems have several potential sources
•Skill scores themselves may be misleading or
of little “real” value
Roads and Maisel
1991 WAF:
MRF has regional
biases in
precipitation over
long periods
Example of human improvements on
numerical QPF (Olson et al. 1995, WAF)


   NGM        Manual     OBS
Slow improvement
in skill for human
forecasters, but less
skill for heavier
amounts (Olson et
al. 1995, WAF)
Annual bias has
also improved
slowly, but
interestingly, is
better for Day 2
than Day 1 (Olson
et al. 1995, WAF)
QPF skill is better
is winter than in
summer, even
when forecasters
adjust the NWP
guidance
   What are sources of
      QPF error?
• Resolution inadequacies
• Parameterization errors
• Initialization deficiencies
• Observational errors in
  verification
    If vertical motion is directly
      constrained by horizontal
            resolution…..
•Shouldn’t forecasts for heavy rain events be
   greatly improved with finer resolution?
 •Is there a “magic” resolution where model
  QPF will approach observed peak values
Gallus 1999 found QPF-horizontal resolution
dependence is case-dependent and varies with
convective parameterization
          6/16/96          6/14/98              7/28/97
        Mx obs: 225     Mx obs: 330          Mx obs: 250




        7/17/96          5/27/97
                         Mx obs: 102
          Mx obs: 300

                                       BMJ -shaded
                                       KF - clear
Extreme
example of
unexpected
results and
Conv. Param.
Impacts:
7/17/96 00UTC
surface
conditions
00 UTC 17 JUL 1996 -
OMAHA

Betts-Miller- Janjic
Reference T, Td
profiles shown
Large MCS
drops up to
300 mm of
rain, causing
record river
crests and
severe flash
flooding in far
eastern NE
and western
IA.
7/17/96 BMJ
simulations with
78,39,22 and 12 km
horizontal resolution
                           MX: 46    MX: 45




NOTE: actual
reduction in peak QPF
amounts as resolution
improves                MX: 32      MX: 32
7/17/96 KF
simulations:

                            MX: 11      MX: 70
NOTE: very strong
QPF sensitivity to
horizontal resolution.
Precipitation area
shifted much farther     MX: 135
                                     MX: 186
north than in BMJ
runs, or observations
Daytime precipitation (12-00 UTC 7/16-17/96)
BMJ produces much larger area and amounts
       BMJ                             KF
Convective scheme influences cold pool
strength, which in turn, affects evolution of
events outside initial rain region
Impacts of convective schemes may be felt outside
region of precipitation.
Here, stronger downdrafts in KF scheme result in
greater northward transport of instability into
Minnesota - leading to more intense subsequent
development.




           BMJ                     KF
   Another case: Iowa flood of
           June 1996
   Large-scale region looked favorable for
               excessive rains
Heaviest rains (225 mm) fell in small area in
                 warm sector
  Impacts of horizontal resolution changes
strongly depend on convective scheme used
Tropical-like soundings
with very deep moisture
Td at 850 mb = 18 C
Td at 700 mb = 8 C
BMJ simulations:
Almost no
horizontal
resolution-QPF
dependence
No hint of C IA
maximum
12 UTC 6/16 cold pool affecting Iowa
12 UTC 6/16
Eta model 00 hr
- initialization
NOTE: cold
pool is missing:
winds are
southerly,
without E
component
21 UTC 6/16
Observed
Surface Moisture
Convergence


Flood-producing
storms would
form on C IA
enhancement
Simulated Moisture
Convergence -21
UTC - BMJ run with
12 km resolution
Despite poor initial
wind field, model
does show
enhancement in W IA
BMJ simulation:
No general clearing
into Iowa by 1 pm -
Less destabilization
than actually
occurred
KF simulations:
Strong horizontal
resolution-QPF
dependence
Some evidence of C
IA enhancement
with 22 and 12 km
resolution
KF 6 hr
forecast:
Some
clearing into
SW Iowa
more
agreement
with obs.
        June case shows:
• Moist low-mid troposphere allows BMJ
  scheme to be aggressive
• Even high resolution may not improve
  simulation of small QPF maxima if other
  simulated parameters are incorrect
• Generation of QPF upstream due to
  resolution changes may affect QPF
  downstream
 Changes within a specific
convective parameterization
    can also have a very
 pronounced effect on QPF
  Spencer and Stensrud (1998) show
   this using MM4 with KF scheme
  Spencer & Stensrud variations in
           KF scheme


•Permit Precipitation Efficiency to
remain at maximum (90%) instead of
varying from 10-90%
•Neglect convective downdrafts
•Delay convective downdrafts
   Max. QPF for 4 MM4
  MaximumPrec in 4 KF tests runs




From Spencer and Stensrud 1998 - MWR
Microphysical schemes may be the
 next challenge - once resolution
   improves so that convective
  parameterization is no longer
            necessary
 • Colle and Mass examine resolution-
   orographic precipitation (1999) dependence
 • Microphysical schemes influence results
OBS PRECIP IN
PACIFIC
NORTHWEST
FLOOD EVENT
(1996)
from Colle and Mass
(1999; MWR)


Pronounced
orographic effects
4 km MM5 run does well at crest but
underestimates lee precipitation
       Horizontal resolution affects
       precipitation patterns near
       mountain due to resolution of
1.33   mountain wave effects. Model
       QPF performance in lee of
  4    mountain fluctuates - low bias
       is best in coarsest run, but
       heaviest precipitation just to lee
 12
       of crest occurs with highest
 36    resolution
Although
precipitation
forecasts generally
improved as
resolution was
refined from 36 to
4km, little additional
improvement
occurred with 1.3
km resolution (Colle
& Mass)
Model QPF in
relation to
resolution of
topography
Microphysical
schemes may have
significant influences
at high resolution.
Colle and Mass (1999;
MWR) found that lee-
side precipitation was
too small in high-res
MM5 simulations,
partly because snow
fallspeeds were too
large.
Best results may not occur with most sophisticated
microphysical scheme
Microphysical scheme differences affect QPF in
different areas
Mesoscale initialization may be
    poor and affect QPF



    Stensrud and Fritsch (1994) have
  shown the impacts of improved cold
           pool initialization
         Typical initialization
             deficiencies

•   Low-level jet characteristics
•   Cloud boundaries
•   Fronts and drylines
•   Convective outflows
•   Surface characteristics
            Stensrud and
            Fritsch 1994
            MWR:


            Initialization of
            NE KS mesoscale
            boundary has
            important impact
MM4 -25KM
            on QPF
      How do we verify QPF?
• Bias scores (how many grid points have X
  amount of rain compared to observations)
• Threat Scores (area correct/(area forecast+
  area observed - area correct))
• Probability of Detection
PRIMARY VERIFICATION
TOOLS TODAY USED BY
  NCEP FOR QPF ARE:
   BIAS: Number of grid points having
  simulated rain of X amount divided by
number of observed points with X amount
EQUITABLE THREAT SCORE: Ratio of
 correct forecasts (hits) to total forecasts +
  observations - hits (with correction for
                chance hits)
                    BIAS
•B=F/O
•Can vary from 0 to >> 1
•Bias > 1 means the model is generous with areal
coverage of precipitation
•Bias < 1 means the model doesn’t generate
enough areas with precipitation
•Many operational models have B>1 for small
precipitation amounts, and B<1 for large amounts
                    ETS

•ETS=(H-C)/[F+O-(H+C)]
•0<ETS<1
•Similar to a Threat Score but takes into account
that even “chance” forecasts will be correct some
of the time (Schaefer 1990; Gandin and Murphy
1992)
              1995-1997 ETS AT NCEP
                  (Mesinger 1998)
        .01    .10 .25   .50   .75   1.00 1.50 2.00
29ETA



48ETA



MRF



NGM
     How valuable are these
     verification methods?
 • Model A covers your state with 1inch of rain
 •Model B simply produces 5 inches in the one
              county to your east
•A lone supercell drops 5 inches on your county


     Which model had the
     better forecast?
   What are our Bias and ETS?
For measurable precip (or any category less
               than 1 inch):
Assume one grid point per county with 100
             counties in state
     Bias in model A: 100/1 = 100.0
       Bias in model B: 1/1 = 1.00
      ETS in A: 1/(100+1-1) = 0.01
        ETS in B: 0/(1+1-0)= 0.0
    Objective scores may not
    agree with your answer!
   Improved mesoscale QPF verification may
      involve a phase shift of the simulated
precipitation field. Kalnay and others (1999) are
            studying such an approach
      Concluding Thoughts
•QPF is probably the most difficult aspect of NWP
- the hardest one to envision being solved in 25
years
•If convective parameterizations are used, behavior
of these schemes exerts powerful impact (primary
differences between different models are probably
related to the Cu scheme)
•Thus, forecasters can benefit by understanding the
specifics of how the schemes behave
   Concluding Thoughts (Cont.)
• At very high resolutions, microphysics will
  likewise complicate the picture
• Forecasters need to be aware of small-scale
  boundaries of importance, which will most
  likely be poorly depicted in initialization
• New methods of evaluating what is a
  “good” QPF will be needed

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:0
posted:5/6/2014
language:Unknown
pages:60