Improvement in model predictability in the monsoon area of S. America: impact of a simple super-model ensemble

1
Improvement in model predictability in the
monsoon area of S. America impact of a simple
super-model ensemble Pedro L. Silva Dias
Demerval S. Moreira. Institute of Astronomy,
Geophysics and Atmospheric Sciences
University of São Paulo VAMOS VPM8 Modeling
Workshop Mexico City, 09 to 11 March 2005
2
THORPEXA Global Atmospheric Research
Programmewww.wmo.int/thorpex
3
Resumé of Science Plan

• Research on weather forecasts from 1 to 14 days
  lead time
• Four research Sub-programmes
• Predictability and dynamical processes
• Observing systems
• Data assimilation and observing strategies
• Societal and economic applications
• Emphasis on ensemble prediction
• Interactive forecast systems tuned for end
  users e.g. targeted observations and DA
• THORPEX Interactive Grand Global Ensemble
• Emphasis on global-to-regional influences on
  weather forecast skill

4
SALLJEX Intercomparison Program 2003 GEF
Evaluation of Numerical Forecasts available in
the Plata Basin December 2004
5
Operational NWP and NCP at CPTEC

• Weather Forecasting Operational Suite (black
  2003red2004)
• Global Spectral Model T 215L42 up to 7 days, two
  times a day NCEP analysis, GPSAS/DAO assimilation
  (6 hours)
• Regional Eta Model (40) - 20kmL38, up to 5 days,
  two times a day RPSAS/DAO CPTEC regional analysis
  CPTEC global model BC
• Global Ensemble T126L28, up to 15 days, twice a
  day, 15 membersCPTEC/FSU ensemble principal
  components scheme
•  

6

• Seasonal Prediction
• Global Spectral Model T062L28 up to 4-6 months,
  once a month
• 25 members each IRI mode (anomaly based on
  (10) 50 years)
• now CPTEC is an IRI member
• running two more sets of seasonal forecasting
  
• DERF mode
• and alternative Cu Parameterization
• Boundary conditions
• Monthly SST persisted anomaly (observed) or
• predicted (Tropical
  Atlantic (statistical) and Tropical
• Pacific)
• Initial climatological
  values soil moisture
• albedo and snow depth

7
Institutições com atividade em modelagem/previsão
Meteorológica Hidrológica
Investigação/ Operacional Univ. Federal do Rio
de Janeiro Universidade de São Paulo Fundação
Universidade do Rio Grande do Sul CIMA
INMET
UFRJ
USP
CPTEC
SIMEPAR
Operacional/Pesquisa Centro de Previsão de Tempo
e Estudos Climáticos
UFSC
FURGS
CIMA
SMA
Serviço Nacional INMET - Brasil SMA - Argentina
8
Instituto Nacional de Meteorologia INMET
Brasil Modelo Meteorológico Sistema de
Assimilação de dados Divulgação
9
http//www.inmet.gov.br/

• MBAR Installed by the German weather service
  (DWD) through WMO agreement in 1999 ()
• 25km resolution, hydrostatic , 310 by 310 points
• Run twice a day 00 and 12 GMT
• Uses boundary conditions from DWD global model
  (internet)
• FORTRAN90 modular
• SGI cluster limited parallelization (12
  processors)
• INMET has 80 processors
• Data assimilation limited to conventional data
  update of DWD analysis

• Large number of products available in real time

• () Also runs at the Directorate for Hydrography
  and Navigation (DHN) - Brazil

10
Servicio Meteorologico Argentino SMN Buenos
Aires - Argentina

• ETA SMN, fue obtenida en el International Center
  for Theorietical Physics, Trieste, Italia y
  adaptada para el extremo sur de Sudamérica por el
  Grupo de Modelado Numérico del Departamento de
  Procesos Automatizados del Servicio Meteorológico
  Nacional.
• Abarca el área definida entre 14 y 65º latitud
  Sur y 30 y 91º longitud Oeste, y utiliza como
  campo inicial y de borde los análisis y
  pronósticos cada 12 horas producidos por el
  modelo global GFS (NCEP).
• ETA SMN pronostica a 120 horas a intervalos de 3
  horas para 38 niveles de presión en la vertical
  con una resolución horizontal de 0.25º.
• El modelo corre en una Origin 2000 (sgi) con 7
  procesadores R10000 en paralelo. Las salidas
  están disponibles dos veces al día y corresponden
  a las corridas de 00Z y 12Z

11
Laboratório MASTER - Universidade de São Paulo
São Paulo SP Brasil

• BRAMS - Brazilian Regional Atmospheric Modelling
  System (RAMS) - version of RAMS (CSU/ATMET)
  partnership since 1989 with FINEP/FAPESP
  support.
• Air pollution module (urban and biomass burning)/
  photochemistry of ozone, convective
  parameterization and transport,surface processes,
  dynamical vegetation validation studies with
  field experiments.
• Weather forecasting up to 3 days, 20km
  resolution, 2X/day BC from CPTEC or NCEP
• Surface data assimilation cycle
• PC Cluster 18 processadores PC (aprox. 2 h)
• Downscaling of the CPTEC seasonal prediction 3
  mo (2-3 members/month)
• Operational System implemented at other
  institutions (FURGS and SIMEPAR)
• Validation against surface metrics

12
SIMEPAR Sistema Meteorológico do Paraná
Curitiba/PR Brasil www.simepar.br

• BRAMS 16 proc. PC-Cluster
• ARPS Origin 2000 16 processors
• Surface data assimilation cycle
• Nesting op. system 64 km and 16 km resolution
• Products not available in public homepage

13
LPM - Universidade Federal do Rio de Janeiro Rio
de Janeiro RJ http//www.lpm.meteoro.ufrj.br/ -
SIMERJ (Meteorological System of the State of
Rio de Janeiro)

• Model MM5 and BRAMS 2 grades configuração de
  30km e 10 km 2 X/dia 00 e 12 GMT
• BC and IC from AVN/NCEP
• Data assimilation not in operational work but
  experimenting with MM5 system.
• Products for the Civil Defense and available in
  open homepage

14
Universidade Federal de Santa Catarina
Florianópolis/SC Brasil http//www.eps.ufsc.br/s
ervico/meteoro.htm

• Model ARPS.
• FORTRAN-90.
• ARPS configured with 3 nested grids based on AVN
  IC and BC (NCEP)
• 60 hour forecast at 40 and 12 km e up to 36 hr
  with 4 km, 2X day.
• PC Cluster PC 14 processors

15
Fundação Universidade Federal de Rio Grande -
Rio Grande/RS

• BRAMS 64 km, 16 km e pequena grade de 4 km
  sobre Porto Alegre
• 60 horas 2X/dia
• Condição inicial e de fronteira do CPTEC
• Não assimila dados de superfície ou altitude
• Cluster de 32 processadores PC

http//www.gepra.furg.br/
16
Centro de Investigaciones del Mar y la Atmósfera
- CIMA Buenos Aires - Argentina

• Versión adaptada en el CIMA del Limited Area Hibu
  Model, con los paquetes físicos del Geophysical
  Fluid Dynamics Laboratory -Orlanski y Katzfey,
  1987)
• La resolución horizontal es de 65 km. en cada
  dirección) y la vertical es de 18 niveles up to
  10mb.
• 2 veces/dia 00 y 12 GMT from NCEP analysis
• Este diseño requiere aproximadamente de 4 horas
  en una SGI-Indigo 2 para completar un pronóstico
  a 72 horas. Este sistema de pronóstico se
  encuentra funcionando en forma experimental desde
  Agosto de 1998.
• Malla E de Arakawa (1972) horiz. Y coordenada
  sigma vert.

17
The Eta model
Settings Large domain for seasonal
simulations Intermediate domain for routine
daily runs Higher resolution (22 km) domain for
studies of hydrologic impacts
72 hr forecasts -
- Initial and boundary conditions AVN NCEP
Reanalyses - Further online information and
forecasts http//www.atmos.umd.edu/berbery/etasa
m
18

• Other models
• FURNAS Belo Horizonte MG Brasil MM5 15 km
  (CI e CF do AVN) operational for internal
  purposes (partnership with UFRJ).
• Serviço Meteorológico de Paraguay WRF installed
  by a private consultant (off the shelve)- (
  operational problems not yet fully
  operational)
• National Laboratory of Scientific Computation
  Petrópolis RJ. Model ETA-Workstation 10km
  research and operation for local civil defense.
• Universidade do Chile Santiago Modelo MM5 (CI
  e CF do AVN) http//www.dgf.uchile.cl/rgarreau/M
  M5/

19
(No Transcript)
20
Integration of models Concept of Super Model
Ensemble

• Several models are available
• global, (CPTEC, NCEP, JMA, ECMWF, UKMO, CMS
  etc)
• Regional models in S. America CPTEC(ETA), INMET
  (DWD), MASTER (BRAMS), SIMEPAR (ARPS, BRAMS),
  UFRJ (MM5, RAMS), UFSC(ARPS), FURGS (BRAMS),
  CEMIG (MM5), LNCC (ETA), UBA (ETA, LMD, RAMS),
  Univ. Chile (MM5), aprox. 14 models !

• Differences in physical processes
  parameterization, data assimilation, data source
  

21

• Brazilian Marine Services
• NCEP
• To be included ECMWF, JMA, BMRC, UKMO
• Project financed by FINEP/Brazil (BRAMSNET).

22
How can we combine several forecasts in an
optimal way???

• Simple solution based on concepts of data
  assimilation

23
Data assimilation the art of inventing
dataObjective combine a forecast with
observations. First step is to perform a
forecast of , e.g., temperature Tb (initial
guess) and then an observation is combined To
through the optimization problem based on the
cost function
The analysis is given by
where and the variance of the
analysis error is smaller than the forecast and
the observation
24
Optimal Forecast
T ? (Ti-Bi)/MSEi
Where Ti is the forecast provided by the
ith model Bi is the ith model
bias MSEi is the ith model
mean square error
25

• Problem
• Bias and MSE need an averaging period
• How long?
• 2 years??? typical sample for MOS

• Practical choice 10, 15, 20, 30 days?

• Intraseasonal signal in model bias suggests
  shorter period

26
Multi model Ensemble Homepage at the MASTER
Laboratory/University of São Paulo
Choose the model RAMSC_25km_/MASTER-Univ.Sao
Paulo (init. CPTEC ) RAMSA_25km_/MASTER-Univ.São
Paulo (init. AVN) RAMSP_25km_/MASTER-Univ.São
Paulo(init. with assimilation cycle) CATT-BRAMS_40
km_g2/CPTEC CATT-BRAMS_20km_g3/CPTEC ETA_40km/CPTE
C (init. CPTEC global) ETA_20km/CPTEC (init.
CPTEC global) ETA_40km/CPTEC (regional
assimilation cycle) ETA-80km_Workstation Univ.
of Maryland ETA_17km_SE_Workstation
CATO/LNCC ETA_10km_LNRJ_Workstation
CATO/LNCC MM5_30km_g1/LPM-Fed.Univ.Rio de
Janeiro MM5_10km_g2/LPM-Fed.Univ. Rio de
Janeiro HRM_30km_DWD regional model at Brazilian
Hydrographic Center MRF/NCEP-global AVN/NCEP-globa
l CPTEC_T126-global Mean CPTEC ensemble_T126/CPTEC
Mean NCEP Ensemble PSTAT (Optimal combination
of all forecasts)
27
(No Transcript)
28
(No Transcript)
29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
(No Transcript)
33
(No Transcript)
34
(No Transcript)
35
(No Transcript)
36
Conclusions

• Simple procedure based on data assimilation
  principles quite successful
• Future optimal choice of the averaging period
  for computing bias and MSE
• Include longer time scales impact on model error
  (e.g., interannual)
• Probably 70 of the potential result ? need to
  improve 30 work done so far is 3 of the
  immediate target.
• Collaborative work!!! Quite a progress!!!!