the ncar/atec real-time four-dimensional data assimilation and forecast system (rtfdda)

Post on 26-Jan-2016

32 Views

Category:

Documents

1 Downloads

Preview:

Click to see full reader

DESCRIPTION

The NCAR/ATEC Real-Time Four-Dimensional Data Assimilation and Forecast System (RTFDDA) Yubao Liu, Laurie Carson, Francois Vandenberghe Chris Davis, Mei Xu, Rong Sheu, Al Bourgeios, Fei Chen and Daran Rife Project Leaders: Scott Swerdlin and Tom Warner. Problems, solutions and goals - PowerPoint PPT Presentation

TRANSCRIPT

The NCAR/ATEC Real-Time Four-Dimensional Data Assimilation and Forecast System (RTFDDA)

Yubao Liu, Laurie Carson, Francois VandenbergheChris Davis, Mei Xu, Rong Sheu, Al Bourgeios, Fei Chen and Daran Rife

Project Leaders: Scott Swerdlin and Tom Warner

Problems, solutions and goals Scientific design Engineering aspects Hardcore issues: experience On-going developments

Issues for Mesoscale Analysis and Forecast

Data are sparse and irregular in space and time and they are not sufficient to describe the structures of local-scale circulations.

Local circulations are complicated – Large-scale forcing and multiscale interaction– Local terrain forcing– Contrasts in surface heating/cooling – Land-soil moisture and thermal properties

A full-physics mesoscale model with accurate local forcing + use of all data.

NCAR/ATEC RTFDDA Is Such a System

• PSU/NCAR MM5 (version 3.6) based, • Real-time and Relocatable, • Multi-scale: meso- meso-x = 0.5 – 45 km

• Rapid-Cycling: at a flexible interval of 1 – 12 hours,

• FDDA: 4-D continuous data assimilation, and

• Forecast ( 0 – 48 hours) systems.

Main Objective: effectively combines the full-physics MM5 model with all available observations to produce best-possible real-time local-scale analyses and 0 – 48 hour forecasts

Main Objective: effectively combines the full-physics MM5 model with all available observations to produce best-possible real-time local-scale analyses and 0 – 48 hour forecasts

Data Assimilation and Forecasting

FDDA is based on “Observation-Nudging” TechniqueStauffer and Seaman (1995), and Numerous modifications and refinements by

NCAR/ATEC modelers.

( See ~20 pubs at https://4dwx.org/publications/ )

FDDA is based on “Observation-Nudging” TechniqueStauffer and Seaman (1995), and Numerous modifications and refinements by

NCAR/ATEC modelers.

( See ~20 pubs at https://4dwx.org/publications/ )

Cold Start t

Forecasts (MM5/WRF)

FDDA (MM5)

Observations (synoptic/asynoptic)

(once a week)

Obs-nudging: Weighting Functions

W = Wqf Whorizontal Wvertical Wtime

Obs-nudging: Weighting Functions

Weighting functions should depend on grid sizes; local terrain; observation location, time, quality, platforms; and air stream properties.

W = Wqf Whorizontal Wvertical Wtime W = Wqf Whorizontal Wvertical Wtime

OBS

OBS

Hi

sfc

Advantages of Continuous OBS-Nudging

Allows for model-defined solution in data-sparse regions, but adjusts for observations where they exist.

Combines the dynamic balance and physical forcing of a model, with observation information available at and before forecast time.

Provide 4-D, continuous, “spun-up” and complete analyses and I.C. for nowcasts/forecasts:– Local circulations and cloud and precipitation fields

• Note: “Analysis Nudging” technique may not be applicable in meso- and scale models

Operational RTFDDA Systems

Regular Operational RTFDDA Systems

Oct.10,00

June.1,02

Feb.5,02

Jul.2,01

Sep.4,01

5 permanent + 8 short-term systems

Special-operation Sites

Afghanistan

Athens-2004Iraq

RTFDDA Engineering Design• MACs and DACs• MACs: 16 – 48-node linux

clusters • 1 - 3 GHz dual-CPUs with

Myrinet networking• Parallel data collection, data

assimilation and post-processing

• Graphics and file service of various formats

• Archiving, verification and local-scale climotologies

• Tools for system monitoring and recovery

Domain Configuration Example

(130 km x 85 km)

Grid 1: 36km Grid 2: 12km Grid 3: 4.0km Grid 4: 1.33 km

2002 SLC Olympics

Diverse and Frequent Observations(An example)

U (m/s)

Modelvs

Observation

(750–300hPa)

Valid at 00 UTC of the 5-day simulation

Soundings (1665)

Profilers (3717)

ACARS (4220)

SATWINDs (2323)

obs

model

BIAS = 0.6

RMSE=3.1

BIAS = 1.1

RMSE=3.3

BIAS = 0.5

RMSE=2.1

BIAS = 1.1

RMSE=4.1

RT-FDDA Model System

GMOD GUI Ensemble Anal/Fcst WRF FDDA

Model physics

Data assimilation schemes

SST, snow cover/depth, sea ice

LSM DA for soil properties

GPS

Sat Tb …

QC

METAR, spec, buoy, ship, temp, pilot, speci, Mesowest, Satellite winds, ACARS, NPN profilers, CAP profilers, radar data, range SAMS, soundings and profilers, cloud / precipitation, and …

ForecastsRTFDDA Analyses

More data; No bad data; Use of data quality Dealing with the model errors – physics biasFine-tuning data assimilation weighting functionConsidering small-scale uncertainties

RUC

ETA..

GMOD (Global Meteorology on Demand)

Summary• The goal of the RTFDDA system is to produce local-

scale four-dimensional analyses and forecasts for various weather-critical applications, tests and events.

• The RTFDDA system has proven to be reliable, reasonably accurate, and widely applicable.

• The operational RTFDDA systems have become a dependable tool for our users.

• Continuous enhancements are being made to improve each system component, including data handling, data assimilation schemes, model physics…

Weather Systems for Regional NWP

Encompass several meteorological scales:– Synoptic ~ 1000 km

• High/low pressure systems, fronts, cyclones …

• Need 20 – 30-km grids

– Mesoscale phenomena ~ 5 – 100 km• Mountain/valley circulations, sea breezes,

convective systems, urban effects …

• Need 0.5 – 10-km grids

The end. Thank you!

Scientific Issues for mesoscale prediction

Predictability-limits favor shorter forecastsStrong dependence on initial conditions (IC)

“Spin-up” of dynamics and cloud/precipitation Sparse and uneven observationsDependency on larger scale models through

boundary conditions (BC)Model physics are extremely important

Solution and GoalsReal-time observations

Collect as many observations as possible Full use of the observations

Conduct dynamic and cloud/precipitation initialization to eliminate/mitigate the “spin-up” problem,

Aim for best-possible “CURRENT” analyses and 0 – 12h forecasts, Predictability is good, Analysis/observations are effective, Large scale model (B.C.) is accurate, Fill the “Gaps” of the national operational center models.

and accurate 12 – 48 hour forecasts.

top related