Rainfall estimation by BMRC C-Pol radar

National S&T Center for Disaster ReductionNational S&T Center for Disaster Reduction

Rainfall estimation by BMRC Rainfall estimation by BMRC C-Pol radarC-Pol radar

ICMCS-V 2006.11.03

1Lei FengLei Feng and 1,2Ben Jong-Dao JouBen Jong-Dao Jou

( 鳳雷 ) ( 周仲島 )　 .

1National S&T Center for Disaster Reduction 2National Taiwan University


ObjectivesObjectives

• To illustrate the ability of rainfall estimation using Areal R(ΦDP) and R(KDP) by BMRC C-Pol radar. Radar-Raingauge comparisons in three different sizes of area :

– Multi-beam Multi-beam (Area ~100 km2) Areal R(ΦDP)

– Single-beamSingle-beam (Area ~ 25 km2) Areal R(ΦDP)

– PointPoint (Area ~ 2 km2) R(KDP)

• Try to correct the wind drift effectwind drift effect when comparing with single raingauge.


2

1

2

1

),(y

y

x

xdxdyyxRAR

2

1

2

1

)],([y

y

x

x

bDP dxdyyxKc

n

j

bjDPjDP

bj rrrr

rrc)],(),([)](2[

2

)(

2 121

1212

NSSL Ryzhkov and Zrnic(1998)

CSU Bringi (2001)

j

n

ji

m

ijiDPjDPjDP rrrrrr

c

),(),(),(

2 1122

2

1

2

1

)],([y

y

x

x DP dxdyyxKcAR

Two Areal Rainfall schemesTwo Areal Rainfall schemes

Notice the difference:Gate area ↑ with range ↑ but NSSL scheme without area weighting

Keep the area weighting, but need ΦDP information at each range gate

1


-30

-35

-40

-45

-5015 20 25 30 35

C-Pol radar at (0,0)

In Darwin

18 rain gauges in the 10 x 10 km2 area

BMRC C-PolBMRC C-Pol rain gaugerain gauge networknetwork

5 single-beams, the area of 5 single-beams, the area of each beam is ~ 25 kmeach beam is ~ 25 km22

1 multi-beam, the area of 1 multi-beam, the area of each beam is ~ 100 kmeach beam is ~ 100 km22

18 raingauges, the radar c18 raingauges, the radar coverage of each gauge is ~ overage of each gauge is ~ 2 km2 km2 2 (radius 0.8 km)(radius 0.8 km)

RD-69


Case A - 15 Jan 1999

Case A, Time series plot (100 km2)


Case B - 01 Mar 1999

Case B, Time series plot (100 km2)


Case C - 17 Mar 1999

Case C, Time series plot (100 km2)


Multi-beam results

• Area size: ~100 km2

• Very high correlation coefficient: 0.97

• Small standard deviation: 1.99 mm/hr

• Little underestimate• Sample number: 108

case (A+B+C)


Single-beam results


• High correlation coefficient: 0.94

• Small standard deviation: 3.43 mm/hr

• Little underestimate• Sample number: 590

case (A+B+C)


Point results


• Low correlation coefficient: 0.86

• Large standard deviation: 6.38 mm/hr

• Under estimation• Sample number: 1091

case (A+B+C)

• The result is getting worse as the verification area getting smaller. Why ?


Point Comparison ProblemsPoint Comparison Problems• Inherence difference of the measurements: Rain

gauge accumulates continuously rainfall on a point while radar samples almost instantaneously a volume averaged rainfall rate.

2

2

2

22

),,(1

),,(L

L

L

Lradar dxdytvyuxRL

tyxR

• Zawadzki (1975) already described Radar-Gauge comparison problems:

2

2

),,(1

),,(t

tgauge dttyxRt

tyxR

Wind drift effect

Time lag effect


Can we correct the wind drift effect ?Can we correct the wind drift effect ?

from DLOC

Strong horizontal wind Overestimate or Underestimate ?

How about the wind drift effect ?How about the wind drift effect ?


OptimalOptimal offsetoffset vectorvector

• An area of radar data which covering all surface rain gauges is moved around the original point in a square window (8km x 8km) with 200 m interval in X and Y direction.

• The cross-correlation coefficient is calculated between the time lagged (1.5 minute) surface rain rates of the gauges and the space shifted radar rain rates.

• A two-dimensional correlation field is produced. The distance from the point of the maximum correlation to the original point was defined as “optimal offset” of the horizontal displacement.


OptimalOptimal offset vectoroffset vector

National S&T Center for Disaster ReductionNational S&T Center for Disaster Reductionoptimal offset vector

-4

-2

0

2

4

-4 -2 0 2 4

X(km)

Y(k

m)

15-J an

1-Mar

17-Mar

Only 39/89 volumes can be easily found out the offset vectors, most of them are convective type rain.

Case B

Case A

Case C

Is these two reasonable ?

2 km


Checking the optimal vector far from system moving velocity case


No wind drift correctionAfter wind drift correction

Case A Case B


If the coefficient of R(KDP) estimator increase 50%, it look better.

Can we do this change for this case ?

No wind drift correctionafter wind drift correctionBut significant underestimation


Rainfall with smaller raindrops need to use higher coefficient in R(KDP) estimator

Adopted from L. D. Carey ATMO 689

big Zdr ~ big Do

small Zdr ~ small Do


Averaged Volume median diameter using Zdr(D0)

CASE

gauge rain rate radar rain rate

> 5 mm/h >10 mm/h > 20 mm/h > 5 mm/h > 10 mm/h >20 mm/h

A 1.38 1.44 1.54 1.38 1.46 1.55

B 1.42 1.46 1.53 1.44 1.46 1.51

CC 1.09 1.09 1.10 1.10 1.14 1.14 1.11 1.11 1.20 1.20 --

In case C, D0 is significant lower than case A and B


Storm motion


Disdrometer observation Radar estimation

Volume median diameter DVolume median diameter D0 0 estimationestimation

Note: the comparison here is not the same case, but are similar squall line type precipitation in Darwin.


No wind drift correctionAfter wind drift correction


summary (1)summary (1)• It’s very important to consider the wind drift

effect when doing single point radar-gauge comparison. In this study, the normalized error has 17% improvement.

rainrate A rainrate B rainrate C Rainrate ABC Do

No Wind Drift correction 0.78 0.85 0.74 0.81 0.74

Wind drift Wind drift correctioncorrection 0.93 0.93 0.96 0.96 0.87 0.87 0.93 0.93 0.93 0.93

Correlation Coefficient of radar-raingauge comparison

N

igauge

N

igaugeradar

RN

RRNNE

1

1

1

1


summary (2)summary (2)

• Use the BMRC C-Pol radar phase base estimator to estimate rain rate is very accurate , especially on convective rainfall.

• For accurate rain rate estimation, it needs to consider the DSD variability such as stratiform rainfall, orographic rainfall, shallow convective warm rain and so on when using R(KDP) estimator.


Lag 3 min Lag 4 min

Optimal vector Finding, 06:40 15-Jan-1999 (C-Pol at Darwin)

Lag 0 min Lag 1 min Lag 2 min

Lag 5 min

Documents

Rainfall estimation by BMRC C-Pol radar