qpe bob kuligowski noaa/nesdis/star walt petersen nasa-msfc

September 29, 2008 1

QPEQPE

Bob KuligowskiBob KuligowskiNOAA/NESDIS/STARNOAA/NESDIS/STAR

Walt PetersenWalt PetersenNASA-MSFCNASA-MSFC

GOES-R Science Meeting


OutlineOutline

SCaMPR Overview (Bob)SCaMPR Overview (Bob) SCaMPR and Lightning: Previous Work (Bob)SCaMPR and Lightning: Previous Work (Bob) SCaMPR and Lightning: Ongoing and SCaMPR and Lightning: Ongoing and

proposed Work (Walt)proposed Work (Walt) Questions / Discussion (all)Questions / Discussion (all)


SCaMPR OverviewSCaMPR Overview

SSelf-elf-CaCalibrating librating MMultivariate ultivariate PPrecipitation recipitation RRetrievaletrieval Calibrates IR predictors to MW rain ratesCalibrates IR predictors to MW rain rates Updates calibration whenever new target data (MW Updates calibration whenever new target data (MW

rain rates) become availablerain rates) become available Selects both best predictors and calibration Selects both best predictors and calibration

coefficients; thus, predictors used can change in coefficients; thus, predictors used can change in time or spacetime or space

Flexible; can use any inputs or target dataFlexible; can use any inputs or target data


SCaMPR Overview SCaMPR Overview (cont.)(cont.)

Two calibration steps:Two calibration steps:» Rain / no-rain discrimination using discriminant Rain / no-rain discrimination using discriminant

analysisanalysis» Rain rate using stepwise multiple linear regressionRain rate using stepwise multiple linear regression

Classification scheme based on BTD’s (ABI only):Classification scheme based on BTD’s (ABI only):» Type 1 (water cloud): TType 1 (water cloud): T7.347.34<T<T11.211.2 and T and T8.58.5-T-T11.211.2<-0.3<-0.3

» Type 2 (ice cloud): TType 2 (ice cloud): T7.347.34<T<T11.211.2 and T and T8.58.5-T-T11.211.2>-0.3>-0.3

» Type 3 (cold-top convective cloud): TType 3 (cold-top convective cloud): T7.347.34>T>T11.211.2


Predictors from ABI / GLM

Target rain ratesAggregate to spatial resolution of target data

Match raining target pixels with corresponding predictors in

space and time

Match non-raining raining target pixels with corresponding

predictors in space and time

Calibrate rain detection algorithm using discriminant

analysis

Calibrate rain intensity algorithm using regression

Create nonlinear transforms of predictor values

Apply calibration coefficients to independent data

Subsequent ABI / GLM data

Matched non-raining predictors

and targets

Matched raining predictors and

targets

START

END

Ass

embl

e tr

ain

ing

data

set

Cal

ibra

te r

ain

rate

re

trie

val

Out

put


SCaMPR and Lightning: Previous SCaMPR and Lightning: Previous WorkWork

Back in 2005, performed some experiments using Back in 2005, performed some experiments using National Lightning Detection Network (NLDN) data as National Lightning Detection Network (NLDN) data as predictors to SCaMPRpredictors to SCaMPR

Basic predictor was number of lightning flashes in a Basic predictor was number of lightning flashes in a GOES pixel in previous 15 minGOES pixel in previous 15 min

Two uses of the predictor:Two uses of the predictor:» Classification: separate convective (lightning present) Classification: separate convective (lightning present)

and stratiform (no lightning) calibrationsand stratiform (no lightning) calibrations» Lightning flash rate as a predictor for the convective Lightning flash rate as a predictor for the convective

classclass Positive impact; hoping to incorporate into the real-Positive impact; hoping to incorporate into the real-

time version of SCaMPR by year’s endtime version of SCaMPR by year’s end


Impact of Lightning on SCaMPR Impact of Lightning on SCaMPR CorrelationCorrelation

Time Series of SCaMPR Correlation Coefficient vs. MW Rain Rates

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

0.45

0.50

0.55

0.60

0.65

0.70

239.01 240.01 241.01 242.01 243.01 244.01 245.01

JJJ.HH

Co

rre

lati

on

Co

eff

icie

nt

Control Lightning


Impact of Lightning on Impact of Lightning on SCaMPR BiasSCaMPR Bias

Time Series ofSCaMPR Volume Bias Ratio vs. MW Rain Rates

0.1

1.0

10.0

239.01 240.01 241.01 242.01 243.01 244.01 245.01

JJJ.HH

Vo

lum

e B

ias

Ra

tio

Control Lightning

Prediction_SEVIRI_000

No

Loop through pixels

Compute solar zenith angle [POSSOL]

Assign inputs

Adjust reflectances for solar zenith angle

Yes

Determine rain class [Rain_type]

Derive output from inputs and coefficients [gforecast]

GLM Information Content for SCaMPR• Rain Presence• Rain Class/Type (C/S/Other)• Rain Rate (Gross PDF constraints)• IWP (not a requirement)

Rain_type

End Rain_type

Determine pixel latitude band (1-4)

T7.34 < T11.2?

T8.5 - T11.2<-0.3?

Initial Class=1 (water cloud)

Initial Class=2 (ice cloud)

Initial Class=3 (cold-top

convective)Yes

Yes

Determine pixel longitude band (1-4)

Combine pixel latitude and longitude bands with initial class to get final rain class

No

No

SUB

Construct predictor from input variables and coefficients

Loop through predictors

Rain rate predictor?

Use to build rain rateUse to build rain / no

rain discriminator

Rain / no rain discriminator >= threshold?

Rain rate=0

No

Yes

No

No

gforecastSUB

Yes

GOES-R GLM QPE ALGORITHM: Conceptual…………….

Petersen, 9/29/2008

Lightning

Yes

SCaMPR Rain Type SCaMPR GForecast

Lightning-Producing Entity/Weather Type

Location, Cluster, Extent, Characteristics (Flash Rate [FR], D(FR)/Dt, location etc.

Volume Rain Character Ice Volume Rain Character Ice

Convective Fraction

Growing Mature Decaying

Stratiform Fraction

Growing Mature Decaying

GLM, TBD DATA (Boldi, SCaMPR

No SCaMPR

Convective and Stratiform (C/S) Precipitation

Recognized in the precipitation community as a fundamental characteristic of convective system rainfall process since the 70’s-80s

Implemented in most state of the art ground-based radar estimation algorithms (historically problematic for satellite algorithms Vis/IR and MW alike (both fundamental to SCaMPR)

For satellite QPE- this application of lightning information is “low-hanging fruit”: Lightning occurs predominantly in precipitating convection and when lightning does occur, it identifies/pin points convective areas of precipitation.

When combined with other observations- can enhance the classic “presence or absence” problem and facilitate better C/S partitioning (e.g., Grecu and Anagnostou; convective/stratiform fractions and revised rain volume calculation).

Relative to applications in C/S partitioning we can ask (and suspect):

Fundamentally, between raining convective systems of similar area coverage does the presence/absence/amount of lightning identify a systematic difference (e.g., constraint) in system-wide C/S precipitation behavior?


Turn to TRMM and Testbeds: Feature C/S behavior as f(lightning)Turn to TRMM and Testbeds: Feature C/S behavior as f(lightning)Establishing Establishing broadbroad constraints in behavior constraints in behavior

For each TRMM Orbit and precipitation feature:

1. 2A25/2A12 Convective/Stratiform rain fractions (within 1Z99)

2. C/S partitioned Avg./Max rain rate, volume rate, IWP, LWP, Scattering Index

3. Lightning flash count

Evaluate over TN Valley (first)1. C/S feature fractions as f(flash, no flash)2. C/S feature area fain volume rates f(flash,

no flash)3. C/S Rain volume ratios f(flash, no flash)4. Feature C/S Avg. rain rates f(flash, no

flash)5. C/S IWP/LWP f(flash, no flash)

PR LIS

7-Year TN Valley Feature Sample number and C/S Fraction7-Year TN Valley Feature Sample number and C/S Fraction


21,788 Total Features (20 dBZ / 250 K 85 GHz)Feature area > 500 km2

•82 % Had convection•33 % Had lightningAll features (including area <500 km2 )•33% were “convective” (artifacts present)•Only 5% Had lightning

With lightning present there is a clear increase in convective fraction regardless of feature area

This is most pronounced for smallest features (PR/TMI partitioning artifact?)

Average Feature Rain Rate (C/S)Average Feature Rain Rate (C/S)


• Average convective RR ~ 10%-larger for features with lightning regardless of feature area.

• No difference for stratiform. Asymptotic 2 mm/hr stratiform rain rate is similar to Adler and Negri (1988) value.

Convective

Stratiform

C/S Rain VolumeC/S Rain Volume


Total Convective rain volume for features w/lightning is markedly larger regardless of area

Volume rain rate behavior (e.g., slope of line) different for convective feature areas with and without lightning

About the same for stratiform

Future/Ongoing WorkFuture/Ongoing Work


Bias = -10% (-0.99 mm)

Error = 12%

• Calibrated/QC’d dual-pol rain map C/S partitioning with LMA for TN Valley (Proxy dataset?)

• Relate statistics of C/S and flash behavior over lifecycles, compared to TRMM “snapshots”.

• IWP retrieval validation (Not requirement, yet…….).

• Process ASCII features and C/S file for the globe (almost done). Freely available to those who abhor…hate…HDF.

• Extend TRMM features C/S study to global domain including African SEVIRI and other SCaMPR/GLM testbeds.

• Add C/S partition/identifier to Boldi cell ID/tracking algorithm.

• ISOLATE BEST SCaMPR APPLICATION

N. Alabama Testbed TRMM Features/C-S/LIS


Questions / Questions / Discussion?Discussion?

qpe bob kuligowski noaa/nesdis/star walt petersen nasa-msfc

Documents