GRAPES-Based

Nowcasting:

System design and Progress

Jishan Xue, Hongya Liu and Hu Zhijing Chinese Academy of Meteorological Sciences

Toulouse Sept 2005

Outline

• Background• System design• Preliminary results – hydrometeor retrieval and model hot

start

• Further development• Summary

What is GRAPES

Global / Regional Assimilation and Prediction System

Chinese new generation numerical weather prediction system consisting of :

DA, Unified dynamic core, Model physics

Background

Background

Exploit the potentials of GRAPES • To improve the warning of mesoscale severe

weather events in advance of 3-6hr

• To promote the application of remote sensing and in situ data to monitoring meso scale weather systems

• To meet the needs of high quality weather services for Beijing Olympic Games 2008

Motivation

Outline

• Background• System design• Current status• Further development• Summary

System Design

Data input Data Analysis GRAPES-Meso

Extrapolationand forecasting

Display and dissemination

Validation

System Structure

Data Input

• Conventional observation ( RA & Synop ) • AWS• Weather Radar• Satellite• Profiler• Lightning positioning• GPS• Air craft

Data Analysis

Quick look at basic elements:

(Qlable) usage:

Initializing NWP

First Guess of SA, CA

Background of system id and fcst

Surface Analysis(SA) usage:

Initializing NWP

System id and fcst

display

Cloud Analysis

(CA) usage:

NWP hot start

System id and fcst

display

System Design

Quick look at basic elements (Qlabel)

• Based on GRAPES 3DVar

• Observational data: Raob, Synop, Profiler, GPS, Radar(VAD),

• First Guess: Last analysis, NWP

• Spatial resolution ~ 1km

• Update frequency ~ 3hr currently, 1hr later

System Design

Surface Analysis

• Analyzed variables: V10m, T2m, q, ps

• Observational data: Synop, AWS, Qlabel products

• Analysis algorithm: successive correction+variational adjustment

• Spatial resolution ~ 1km

• Update frequency: 3hr now, 1hr later

System Design

Cloud Analysis

• Utilization: model hot start; convective system identification

• Input data: Qlabel products, synop, Radar, satellite

• Resolution ( model grids)

• Analysis procedure:

System Design

Cloud Analysis

• 3-D cloud analysis （ cloud cover 、 cloud top 、 cloud ceiling 、 cloud classification, vertical velocity in cloud ）

• Observational data （ Synop., Aeroplane ， plofiler,radar, satellite ）

• Algorithm: successive correction with variational adjustments

Schematic CA

Model Start OptionsTime-n Time

GRAPES Forecast

CA Analyses

GRAPES Nudging GRAPES Forecast

GRAPES Forecast

Eta

GRAPES LBC for all runs

Dynamically balanced,Cloud-consistent CA

LII

“ Cold Start”

“ Warm Start”

“ Hot Start”

(no CA analysis)

(pre-forecast nudging to a series of CA analyses..)

(Directly using the balanced CA analysis)

Data input Data Analysis GRAPES-Meso

Validation

Current Status

Extrapolation

And forecasting

Display and

Dissemination

Outline

• Background• System design• Preliminary results – hydrometeor retrieval and model hot

start

• Further development• Summary

Retrieval of cloud hydrometeor based on radar observation

Basic assumption:

1, Cloud and rainfall are stationary in short time period and

horizontal advection is negligible.

2, Vertical variation of rainfall is determined by collection

( saturated) and evaporation ( unsaturated) so that the

vertical variations of qc and qv may be derived.

3, In the saturated area the increase of qr is the results of

condensation.

Ⅰ Derive qr from radar reflect factor z

43.1 17.5log( )rZ q ( 43.1) /17.510 /Z

rq

Ⅱ Compute Vt from qr

0.1255.40t rv a q

0.40( )a p p

ⅢⅢ Compute saturation specific humidity

17.27( 273.16)38000035.86exp T

vs p Tq

Ⅳ Compute condensation function

2 2( /1000.0)( )v p vvs

p v vs v

L R c R TgqR c R T q L

F

Ⅴ Ⅴ Compute vertical variation of rain fluxCompute vertical variation of rain flux

( )1 r tq vr zI

Ⅵ Compute qc and qv from rain flux

7 /8c rPRA q q

0.875/( )

v vs

c r r

q q

q I q

0.65( )( )vs v rPRE q q q

0.65

0.0

/( ( ) )

c

v vs r r

q

q q I q

ⅦⅦ Compute vertical velocity in saturated area

/rw I F

Selected case:

2003/07/04 2003/07/04 heavy rain heavy rain

event in Haihe river basinevent in Haihe river basin

Model set up

• Horizontal resolution: 0.04 lat/long• Domain size: 201*201 centered at Hefei city• Vertical layers: 30 with equidistance

• Ztop=15km

• Model Physics : Explicit cloud: Kesller’s Radiation: RRTM for long wave Dudhia for short wave Surface layer:Monin –Obukhov PBL: MRF

• Model initialization:

Cold start: operational analysis

Hot start: qc qr qv wc retrieved

other variables –taken from

operational analysis

dynamic adjustment by “pre-forecast”

model integration

Reflectivity

retrieved qc

Cross section of qc

Prediction of rainfall rate (mm/10min)

Prediction and observation of rain fall

Cross section of cloud and vertical motion

Cloud and precipitation

Further development

• Dynamic adjustment to depress the high

frequency fluctuation due to the unbalance

between cloud-related parameters and large

scale environment;

• Utilization of data of radar net work

• Fusion of radar data with data by satellite and

other equipments

Summary

• A new nowcasting system based on Chinese new generation NWP system and dense mesoscale observational data is being developed;

• The radar data have the potential to retrieve the cloud parameters;

• Model hot start may improve the prediction if the storm is better initialized;

• The problem of unbalance between cloud-related parameters and large scale environment is not solved yet.

Thank you for Thank you for your attention!your attention!