(b) PRV7 vs. Q2 - 2016 Fall Meeting

advertisement
AGU FALL MEETING San Francisco | 3-7 December 2012
Quantification of Spatially Distributed Errors of Precipitation Rates and Types
from the TRMM Precipitation Radar 2A25 (the latest successive V6 and V7) using
NOAA/NMQ over the Lower United States
Sheng
1,2
Chen ,
1,2,3
Kirstetter
,
P.E.
Y.
1,2
Hong ,
J.J.
3
Gourley
, J.
3
Zhang ,
K.
3
Howard ,
J.J.
4
Hu
1School
of Civil Engineering and Environmental Science, University of Oklahoma, OK, USA
2Atmosphere Radar Research Center, University of Oklahoma, Norman, OK, USA
3NOAA/National Severe Storms Laboratory, National Weather Center, Norman, OK, USA
4School of Computer Science, University of Oklahoma, Norman, OK, U.S.A
Results
Introduction
Methodology
Information on the spatial error characteristics of satellite-based
quantitative precipitation estimates(QPEs) are important for
application of satellite rainfall products including weather,
hydrology and climate studies. The uncertainty of the QPEs will
enable the QPE developers improve the QPE algorithms.
In this study, the spatial error structure of surface rain rates and
types from NASA/JAXA Tropical Rainfall Measurement Mission
(TRMM) Precipitation Radar (PR) was systematically studied by
comparing them with NOAA/National Severe Storms
Laboratory’s (NSSL) next generation, high-resolution (1km/5min)
National Mosaic QPE (Q2) over the TRMM-covered Continental
United States (CONUS). The difference between the latest
successive version-6 PR(PRV6) and version-7 PR (PRV7) will be
quantified and mapped around the southern CONUS
Time and location matching technology is applied to obtain the instantaneous matching pairs
of PR vs. Q2(Fig. 3) conditioning on time difference less than 2.5min and the range of field of view of
PR with footprint resolution 5km.
Study Region and data
Study region is southern conterminous United States (CONUS) in
longitude of -124°N to -67°N and latitude 25°N to 37°N. The data
are composed of NOAA/National Severe Storms Laboratory’s
(NSSL) next generation, high-resolution (1km/5min) National
Mosaic QPE (Q2), PRV6 and PRV7 level 2 products 2A25.Data
time spans from Dec. 2009 through Nov. 2010. Q2 was
considered as reference against which PR data were evaluated.
1
(b)
Weight
(a)
(a)
0.75
PR FOV
0.5
-2500
-1000
0
1000
2500
(c)
Q2 grid
Fig.3 (a)PR and Q2 spatial overlapping. (b)Q2 and PR location matching.
The reference mean rainfall
can be computed as:
Rref ( A) 
N
1
N
 i
  Q 2( a )
i
i 1
f
with i 
2
Fig. 4 Robust pairs for PRV6 vs. Q2
( ,  0 )d
(d)Q2 strat.
(a)PRV7 conv.
(e)PRV7 strat.
Fig. 8 (a-c)scatter plots, CDF of occurrence(CDFc) and
(f)
volume(CDFv), and contingency metrics as a function
of different thresholds for PRV6 vs. Q2.(d-f) The same
Fig. 10 (a-c)Total convective rainfall Q2 and PRV7 and scatter plot.
as (a-c) but for PRV7 vs. Q2.
V1

V1  V2
N
  (Q2(a )  R
i 1
i
i
ref
( A))
2
(d-f)Total stratiform rainfall Q2 and PRV7 and scatter plot.
Fig. 5 Robust pairs for PRV7 vs. Q2
(a)
(d)
(g)
(j)
(b)
(e)
(h)
(k)
(c)
(f)
(i)
(l)
N
with V1   i V2   
i 1
 ref
V2

2
V1
i 1
2
i
N
 (Q2(a )  R
i 1
i
ref
( A))
2
Fig. 6 Total precipitation derived from PRV6.
Fig. 11 Spatial Bias(1st row), RB(2nd row),RMSE(3rd row) and CC(4th row) for PRV6, PRV7 and their difference.
Rref   footprint
Conclusion
5000
Fig.2 Join available pairs for (a)PRV6 vs. Q2 and (b) PRV7 vs.
Q2(right.
(a)Q2 conv.
(c)
Robust pairs selection conditioning on:
30
(f)
(c)
i
 mesh ( ai )
Fig.1 (a)TRMM PR satellite; (b)WSD88 Radar locations and TRMM PR
coverage
10000
Fig. 9 (a-c)Total convective rainfall Q2 and PRV6 and scatter plot.
(d-f)Total stratiform rainfall Q2 and PRV6 and scatter plots.
i 1
http://trmm.gsfc.nasa.gov
15000
(e)
(b)
in the domain(A) of PR FOV
Rref
N
(b) PRV7 vs. Q2
(e)PRV6 strat.
(a)PRV6 conv.
(b)
(a) PRV6 vs. Q2
(d)Q2 strat.
gaussmf, P=[0 5]
 footprint
20000
(d)
(a)Q2 conv.
In order to select robust pairs, two weighted standard
errors are computed with the reference rainfall in the
domain(A) of PR FOV , namely:
(a)
Fig. 8 gives the scatter plots, the cumulative density function by occurrence and volume, the
probability of detection(POD), critical success index(CSI) and false alarm ratio(FAR) as a function of
different thresholds for PRV6/7 vs. Q2. Fig. 9 and Fig 10 show the total precipitation from
convective(stratiform) rainfall. Fig 11 shows the Spatial Bias, RMSE and CC for PRV6, PRV7 and their
difference.
Quantitative comparison was carried out with statistics
metrics bias(Bias), relative bias(RB), root mean squared
error (RMSE), and correlation coefficient(CC).
Fig. 7 Total precipitation derived from PRV7.
 PRV7 decreased the underestimation of rain rate from 22.09% to 18.38%.
 PRV7(V6) is moderately correlated with Q2 with a mean CC of 0.58(0.56).
 PRV7 has close CDFc, CDFv, POD,CSI and FAR with PRV6.
 PRV7 and PRV6 shares similar spatial patters of Bias, RB, RMSE and CC with PRV6 over south CONUS.
 PRV7 detected more stratiform precipitation than PRV6 and seen less convective rainfall than PRV6.
Reference:
Contact:
Kirstetter, P. E., et.al (2012), Toward a Framework for Systematic Error Modeling of Spaceborne Precipitation Radar with NOAA/NSSL Ground Radar-based National Mosaic QPE, Journal of Hydrometeorology.
Sheng Chen
Chen, S., et.al (2012), Rates and Types from NASA/JAXA Space-borne Radar against NOAA/NSSL High-resolution Ground-based National Mosaic Radar Network within TRMM-covered Continental of US, Journal of University of Oklahoma
Hydrometeorology(submitted).
Email: chenshengou@ou.edu
Paper No.
H33C-1348
Download