Diabatic Mesomodel Initialization Diabatic Mesomodel Initialization Using LAPS - an effort to generate Using LAPS - an effort to generate

accurate short term QPFaccurate short term QPFBy

John McGinley, NOAA Forecast Systems Lab

With contributors

Steve Albers, John Smart, Paul Schultz FSL BrentShaw, Weather News Intl

And

Gou-Ji Jian, Jen-Hsin Teng, Li-Hui Tai, CWB

OverviewOverview Brief review of LAPS and Hot Start Brief review of LAPS and Hot Start LAPS/ Model Deployments in Sub-tropicsLAPS/ Model Deployments in Sub-tropics

Brief Case StudiesBrief Case Studies VerificationVerification

Ensemble ApplicationsEnsemble Applications Probabilistic QPFProbabilistic QPF

Future WorkFuture Work Blending QPE/QPFBlending QPE/QPF

Our mission at FSL is short range high resolution forecasting. For many customers the first 12 hours is very critical. We needed an approach that allows rapid spin-up of clouds and precipitation.

Useful elements to work with:

1. LAPS cloud and moisture analysis (Water in All Phases: WIAP)2. Variational balancing scheme3. Coupled Mesoscale Models: RAMS, MM5, WRF

The LAPS Diabatic The LAPS Diabatic Initialization TechniqueInitialization Technique

LAPS includes:LAPS includes: Cloud analysisCloud analysis Dynamic balance using variational schemeDynamic balance using variational scheme Coupling to most mesoscale modelsCoupling to most mesoscale models Links to display devices (AWIPS, others)Links to display devices (AWIPS, others)

Sustain the “operational” tradition of LAPSSustain the “operational” tradition of LAPS Robust data ingest, QC, and fusionRobust data ingest, QC, and fusion Platform and model independencePlatform and model independence Computationally inexpensiveComputationally inexpensive

Key element: Water-In-All-Key element: Water-In-All-Phases AnalysisPhases Analysis

Cloud Analysis – satellite, radar, surface, aircraft Cloud Analysis – satellite, radar, surface, aircraft observationsobservations

Water Vapor Analysis – variational methodWater Vapor Analysis – variational method Recovery of microphysical variables using simple cloud Recovery of microphysical variables using simple cloud

model: model: Precipitation Total precipitable waterPrecipitation Total precipitable water Water vapor Integrated liquid water Water vapor Integrated liquid water Cloud water Cloud coverCloud water Cloud cover Cloud Ice Ice Crystals Cloud Ice Ice Crystals

Cloud Analysis SchemeCloud Analysis Scheme

Uses satellite Visible Uses satellite Visible and IRand IR

Aircraft observationsAircraft observations Surface observationsSurface observations RadarRadar Interpolates cloud obs Interpolates cloud obs

to grid with SCMto grid with SCM Cloud feeds back into Cloud feeds back into

water vapor analysiswater vapor analysis

Updated CWB/ FSL scheme (cloud derive subr)

CB

CS

Downdrafts in stratiform region

Less dependence on cloud type,Updraft goes to top of cloud

Randomness in broad convectiveregions

Strongest updraftsin regions of high reflectivity

LAPS Diabatic Initialization

CloudAnalysis

UnivariateData

Fusion

3DVARDynamic

Constraint

LAPSPREP

NWP SystemLAPSPOST

Surface RAOB Sat ACARS GPS Radar(Vr) Profilers Radar(Z) Sat Aircraft METAR

NWP FG

Data Ingest/Quality Control

National NWPLBC

LSM IC

NativeOutput

Forecaster

IsobaricOutput

T, , p, u, v, , RH

T,

qc qi, qr qs, qg

c

T, , p, u, v, RH

Constraints:Mass Continuityu/v Time TendenciesBackground ErrorObservation Error

Adjust for Model:Hydrometeor Concen.Saturation Condition

EquationsEquations

( ) b are background quantities; (^) are solution increments from background; ( )’ are observation differences from background

Dynamic Balancing and Continuity Formalism

Cloud, Wind and Mass Dynamic Adjustment

Cloud, Wind and Mass Dynamic Adjustment

wcFH

FL

T> 0^

q= qs

         

Figure 3. Cloud liquid (shaded), vertical velocity (contours) and cross-section streamlines for analyses (right) and 5-min forecasts (left). The top pair shows LAPS hot-start DI with upward vertical motions where clouds are diagnosed and properly sustained cloud and vertical motions in the forecast; the bottom pair demonstrates the artificial downdraft that usually results from simply injecting cloud liquid into a model initialization without supporting updrafts or saturation. Note that cloud liquid at the top of the updraft shown in the hot-started forecast (above right) has converted to cloud ice.

Hot Start

Cloud onlyCold start

0-Hr 5 min forecast

Balanced Field Unbalanced Field

MODEL NOISE |dp/dt|

Difficult problem: Heavy Rain in the Subtropics Difficult problem: Heavy Rain in the Subtropics - Dominican Republic and Haiti, May 2004- Dominican Republic and Haiti, May 2004

LAPS GUI – Global locatibilityLAPS GUI – Global locatibility

Window MM5 Forecast for Dominican Republic/ Window MM5 Forecast for Dominican Republic/ Haiti Flash Floods 00Z May 22, 2004Haiti Flash Floods 00Z May 22, 2004

TRMM Rainfall EstimatesTRMM Rainfall EstimatesMay 18-25, 2004 May 18-25, 2004 (NASA Goddard)(NASA Goddard)

Taiwan CWB Short Term Forecast System

LAPS (Local Analysis and Prediction System)

Diabatic Initialization technique

Hot-Start MM5

Forecast domains & Computational requirement

D02D03

D01

1km (169*151)

D02

D01

1368 km ( 153 points)

1260

km

( 1

41 p

oint

s)

151

pts

151 pts

9km

3km

CPUs 42 compaq 833 MHz

Need 1.5hrs for 24hrs fcst

0.000.050.100.150.20

0.400.350.300.25

0.710.680.650.620.580.540.500.45

0.920.900.880.860.830.800.770.74

0.990.980.970.960.94

1.00

30 V

erti

cal l

ayer

s (σ

leve

ls)

WRF Run for Typhoon Mindulle using WRF Run for Typhoon Mindulle using

NCEP GFS/NF-15 backgroundsNCEP GFS/NF-15 backgrounds Cold -GFS Hot - NFS

TS .60, .50,.38,.35,.26

TS .53, .51,.44,.29,.26

Verification of Hot Start Forecasts for Tropical Storm Mindulle 0-6 hr forecast - Analyzed gauge data on left; forecast on right

Threat scores for precip categories 1, 10, 20,50, 100, 200mm

6-12 hr forecast - Analyzed gauge data on left; forecast on right

Threat scores for precip categories 1, 10, 20,50, 100, 200mm

6-H Precipitation verification for 4 6-H Precipitation verification for 4 tropical systems in 2003 over Taiwan tropical systems in 2003 over Taiwan

(Jian and McGinley, accepted by JMSJ)(Jian and McGinley, accepted by JMSJ)

○: HOTS

●: NCLDa

Hot Start Cold Start

12-H Precipitation verification for 4 12-H Precipitation verification for 4 tropical systems over Taiwan (Jian and tropical systems over Taiwan (Jian and

McGinley)McGinley)

○: HOTS

●: NCLDbHot Start

Cold Start

Advantage of EnsemblesAdvantage of Ensembles

•Ensembles produce probability forecasts that can be more useful than single deterministic forecasts

•Probabilistic output can be input into economic cost/lost models

• Customers get a “yes-no” forecast based upon skill and spread of ensemble

From WRF/MM5/ RAMS Ensembles, From WRF/MM5/ RAMS Ensembles, to Probabilities, to Probabilities,

to Yes/No Forecaststo Yes/No Forecasts

6-D model grids(variable,x,y,z,t,prob)

Decision Engine

Cost/LossThresholds

Yes or No

Ensemble

Location,Time

Customer

MM5-EtaMM5-Eta MM5-AVNMM5-AVN WRF-AVNWRF-AVN

RAMS-EtaRAMS-Eta RAMS-AVNRAMS-AVN WRF-EtaWRF-Eta

6-Member Ensemble: 3 models, 2 boundary conditions

MM5-EtaMM5-Eta

NN

WRF-EtaWRF-Eta

H

H+1

H+2

H+3

H+4

H+5

Time-Phased Ensemble: an efficient way to get many members in limited computing environments

Time

t0

Ensemble at time = t0 Time weighting is applied to each member

(Number of members) =

(Number of models) x(Length of Forecast) /(Start Interval)

Each pair of runsHas a unique Initial condition basedon LAPS

Radar-observed precipitation -Radar-observed precipitation -24 hrs ending 13 Oct 04 12GMT24 hrs ending 13 Oct 04 12GMT

New approach: time-phased ensemble - model run New approach: time-phased ensemble - model run every hour with new initial conditions - with two every hour with new initial conditions - with two

models, 24 members for every timemodels, 24 members for every time

WRF Time phased ensembleWRF Time phased ensembleModel Run vt 13 Oct 04 18GMT - 18 hr forecastModel Run vt 13 Oct 04 18GMT - 18 hr forecast

WRF Time phased ensembleWRF Time phased ensembleModel Run vt 13 Oct 04 12GMT - 15 hr forecastModel Run vt 13 Oct 04 12GMT - 15 hr forecast

WRF Time phased ensembleWRF Time phased ensembleModel Run vt 13 Oct 04 12GMT - 12 hr forecastModel Run vt 13 Oct 04 12GMT - 12 hr forecast

WRF Time phased ensembleWRF Time phased ensembleModel Run vt 13 Oct 04 12GMT - 09 hr forecastModel Run vt 13 Oct 04 12GMT - 09 hr forecast

WRF Time phased ensembleWRF Time phased ensembleModel Run vt 13 Oct 04 12GMT - 06 hr forecastModel Run vt 13 Oct 04 12GMT - 06 hr forecast

WRF Time phased ensembleWRF Time phased ensembleModel Run vt 13 Oct 04 12GMT - 03 hr forecastModel Run vt 13 Oct 04 12GMT - 03 hr forecast

Ensemble Probabilities:Ensemble Probabilities:Threshold: 5 mm/3 hrThreshold: 5 mm/3 hrat 12 GMT 13 Oct 04at 12 GMT 13 Oct 04

>20

>40

> 60

>80

PrecipitationProbabilities %

Future WorkFuture Work

Ensemble probabilistic post-processing Ensemble probabilistic post-processing and coupling ensembles to decision and coupling ensembles to decision algorithmsalgorithms

Merging QPE and QPF over the 0-6 HR Merging QPE and QPF over the 0-6 HR time frametime frame

Designing a Forecast/ Observation, QPE/ QPF Designing a Forecast/ Observation, QPE/ QPF Blending SchemeBlending Scheme

wi wi wiwi

Forecasts 0H 1H 2H 3H

Observations 0H 1H 2H 3H

CorrelationsCoefficients/

Weights from Training Set

Forecast SetFor New Event

Post-processor

Optimum Forecast

Set

ConclusionsConclusions

1. The LAPS Hot Start Method has shown improvement over operational NWP output In the 0-6 hour time frame.

2. It has demonstrated improved verification of precipitation in higher categories

3. The LAPS /NWP System is capable of running on small computers in local weather offices

4. It has run in winter and summer experiments5. It shows promise for tropical storm QPF when combined with bogussing

the initial position