View
50
Download
0
Category
Tags:
Preview:
DESCRIPTION
The Brazilian Effort on BRAMS and OLAM Pedro L. da Silva Dias LNCC/MCT e IAG/USP Workshop on Weather and Seasonal Climate Modeling at INPE - 08-10 December 2008. RAMS origin: Colorado State University - William Cotton - Greg Tripoli: end 70 ’s early 80’s Cloud Microphysics - PowerPoint PPT Presentation
Citation preview
The Brazilian Effort on BRAMS and OLAM
Pedro L. da Silva Dias
LNCC/MCT e IAG/USP
Workshop on Weather and Seasonal Climate Modeling at INPE -
08-10 December 2008
RAMS origin: Colorado State University -
William Cotton - Greg Tripoli: end 70’s early 80’s
•Cloud Microphysics
•Cloud Dynamics
•Mesoescale Model - Roger Pielke - Virginia University
•Fusion of both models => RAMS (Regional Atmospheric Modeling System)
Mesoscale Modeling at IAG/USP
•1980’s - role of sea-breeze in São Paulo - regional climate - impact of air pollution => lead to fairly complex physics - urban processes, vegetation, topography (numerical challenges)
• Fundamental problem: lack of computer power
•Theoretical studies (80’s ) - instability lines - heat sources
•Late 80’s - Elmar Reiter’s PE model hydrostatic used at CPTEC and USP
•Semi-lagrangean models: remote impact of hurricanes;
•Andes effect: blocking effect - eta coordinate, role of LLJ ‘s
•Computer power limitation -> more emphasis on observational studies;
•RADASP (IPMET-UNESP/USP/INPE): July/81, Jan/Fev 82 ,Jan/89: PBL and convection - mesoscale systems
•ABLE 2a (1985) e 2b (1987) = > atmospheric chemistry - Amazon, biomass burning
•ABRACOS (land use change) - FLUAMAZON ---> LBA (93)
•ABLE is a turning point: beginning of integrated model activities - concept of tracers (radon, CO, aerossols); Beginning of trajectory analysis
•ABLE lead to more observational studies on urban aerosols and urban chemistry in the late 90’s;
•At the end of the 80 ‘s (CONVEX computer at USP - vector and 2 processors);
•Falling behind modeling activities….
•Decision: use a complete model and work on modules:
•MM5, JMA mesoscale model,…RAMS?
•1989: Bill Cotton visits FUNCEME, CPTEC e USP;
•Decision: implement RAMS.
•Strong connection with observational work: model validation (ABRACOS, LBA) and latter a strong connection with urban air quality issues;
•1995 - beginning of regional forecasting - 40km resolution (CPTEC at this time ran ETA at 80km); IBM SP2 with 16 processors in 1997 - boosts operational capability - CPTEC seasonal climate downscaling in 1999.
•Modeling with RAMS: hurricanes, local circulation in São Paulo, São Francisco Valley, NE Brazil, Instability lines in the Amazon, intense cyclones, land use change, impact of pollution sources, convective parameterizations; vegetation (SIB)….
•Paralelism: end of the 90’s -> FINEP project (hardware - PC cluster) - CPTEC role;
•Large number of students - > use of RAMS spread to several universities in Brazil (UFRJ, UFPb, FURGS, UFPA,…)
•FUNCEME begins operational use of RAMS for climate downscaling and weather forecasting
•SIMEPAR - surface data assimilation (FINEP);
FINEP - BRAMS (Brazilian Developments in RAMS)
• Versão 3.0– Based on RAMS 5.04 - ASTER– Maintained by CPTEC– New Functionalities:
• Shallow Cumulus• Deep Cumulus – “Grell-Ensemble”• Soil Moisture Inicialization• SIB2 in addition to LEAF• Surface data assimilation with data quality control• CATT – biomass burning emission module and transport (plus
urban sources).
BRAMSNET• Network of BRAMS users and developers • Inicial partners:
– UFCG– UFRJ– CPTEC– USP– SOMAR– ITAUTEC– FURG
GBRAMSGRID - UFRGS, CPTEC,IAG - climate applications
New implementations:
•Precipitation assimilation - required by some users (SIVAM) - better short range forecasts;
•Urban energy balance and transfer - TEB: required to improve model validation against surface observations;
•Calibration of “Grell Ensemble” with precipitation data;
•New options for the radiative processes in the presence of gases and aerossols - space and time variation - CARMA;
•Interaction cloud/radiation - short wave - (parameterized shallow clouds) - need to improve metric of validation based on fit to surface radiation measurements;
•New options for dry turbulence; (need for improvement of Td diurnal cycle)
•New data assimilation module - based on PSAS/CPTEC
•Simplified photochemistry - 2004 (product of research project with CETESB);
•Full photochemistry module (CPTEC - other presentation)
•Coupling with dynamical vegetation GEMTM - furture IBIS
•Coupling with ocean model (POM) and more recently with mixed layer model
•Coupling with surface hydrology - Sao Francisco, Rio Grande, Uruguai - Pantanal
Code robustness (Jairo Panetta and team):
•Code originally developed by researchers;
•Fundamental rules of software engineering
•Parallel efficiency :
•Challenge: eficiency in vector computers; massive parallelism - shared and distributed memory…
•BRAMS community is growing
• Challenge - Efficient use of High Performance Computing - HPC
• New generation of HPC machines:– architecture
• Massively parallel and vector
– Visualization of of large data sets (3D animation)
– Assistance to “poor mortal “users…
Larger clusters: 1100 processors at CPTEC
NEC SX-6
NEC-SX6 with 12 NEC-SX6 with 12 nodes, 96 nodes, 96 processors in processors in CPTECCPTEC
• Towards Towards Production Code Production Code Effective Effective Portability among Portability among Vector Machines Vector Machines and and Microprocessor-Microprocessor-Based Based ArchitecturesArchitectures
Alvaro Luiz FazendaEduardo Hidenori EnariLuiz Flavio RodriguesJairo Panetta INPE/CPTEC
Contribution of Jairo Panetta - 2007Contribution of Jairo Panetta - 2007
0
100
200
300
400
500
600
700
800
100 200 300 400 500 600
Processadores
Tempo de Execução (s)
Initial
final
What is behind the success of BRAMS?
•Link to observational work!!!!
•Ex. LBA, air pollution programs, micromet tower program
•Operational use for regional forecasting
MRSP
CCN
Poluição de São Paulo - Vôo de 13/08/99 Concentração de O3
0
20
40
60
80
100
120
140
160
12:00 13:00 14:00 15:00 16:00 17:00 18:01
Hora do dia
O3 mixing ratio (ppb)
0
5
10
15
20
25
30
35
40
O3 NO2
PerfilSantos 1.842 m
Cubatão 1.645 m
cpc 13000
PerfilSão Paulo
4440 mTopo
Serra do Mar
1.315 m
São M. Paulista
cpc 18400
Marginalcpc 23500
PerfilSantos 4.276 m
Topo
Measurements with the INPE Bandeirantes aircraft from 11 to 13 August 1999
COO3
CCN
Participação: Inst. Agronômico de Campinas e UNICAMP
Projeto Financiado pela FAPESP: coordenado pelo Dr. H. Rocha (IAG/USP)
Torre da Reserva Jaru –RO - 65m
Rocha 2001
12
3 4
Operational use of BRAMS in regional forecasting
•Note difference between RAMSC (20km) e RAMSB(10km) -
Future of BRAMS
X a
t+ L a X a = N a X
a, X
o, X
v, X
c, X
s+F
aX a ,X o ,X v ,X c ,X s
X o
t+ L o X o = N o X
a, X
o, X
v, X
c, X
s+F
oX a ,X o ,X v ,X c ,X s
X v
t+ L v X v = N v X
a, X
o, X
v, X
c, X
s+F
vX a ,X o ,X v ,X c ,X s
X c
t+ L c X c = N c X
a, X
o, X
v, X
c, X
s+F
cX a ,X o ,X v ,X c ,X s
X s
t+ L s X s = N s X
a, X
o, X
v, X
c, X
s+F
sX a ,X o ,X v ,X c ,X s
X a v l r i , .. . . X o u , v , w , T , sv
, . . .
X v l a ii
, s i gi v
, r o o ti d
, s t o mi c
, V O Ci
, Ci
, Ni
, . . . .
X c C O2
, C H4
, O3
, N Ox
, V O C ' s , S O2
, . . .
X s Ti s
, Wi s
, Ni n
, . . . .
atmosphere
ocean+hydrology + ice
Soil
Vegetation
Gases, aerosols
Modelling Earth System
RAMS/BRAMS is not flux conservative due to:
1. Boussinesq approximation2. Advection operator3. Failure to average divergence over small time-split steps4. Grid nesting applied to primary variables rather than to fluxes
This is not a major problem if lateral boundaries are open
Several users require global domains:
The Ocean-Land-Atmosphere Model (OLAM):
For OLAM global domain, require full conservation
Re-cast governing equations in conservation law form
The Ocean-Land-Atmosphere Model (OLAM): A re-formulation of RAMS for global modeling
Based on a presentation by Robert Walko – Duke University
BRAMS Workshop – May 2006 - CPTEC
OLAM Equations:
V
tv V p 2 OM E G A v g F V
tUx
Vy
Wz
tV F
st
s V F s
p= d R d v R v
CP
CV 1
p0
Rd
CV
Momentum conservation
Mass conservation
Energy conservation
Scalar mass conservation
Equation of state
d
tV v V d p 2 OM E GA v g FV
tV d
tV d F
ts s V d F s.
Discretized equations:
Apply Gauss Divergence Theoremand integrate over Finite Volumes:
tV v V d p 2 OM E GA v g FV
tV d
tV d F
ts s V d F s.
Discretized equations are applied on Cartesian grid with originat Earth center
Grid cell surfaces are not aligned with (x,y,z) coordinates,but each grid cell surface is parallel or perpendicularto local gravity
Numerical algorithm from Wenneker et al. (2002)
RAMS uses terrain-following coordinates
OLAM topography represented byshaved grid cell method
Other features…
• Coded using F90 modules and data structures like RAMS 6.0• Build/compilation procedure same as RAMS 6.0• ‘OLAMIN’ namelist file in same form as RAMSIN• Refined mesh areas specified by location, size, shape in OLAMIN• Global spherical, limited area spherical, limited area cartesian
geometry options• Vertical K at W levels: most natural for evaluation and application• Implicit vertical diffusion solves for fluxes• Implicit surface momentum flux• Graphics done from model itself: No separate REVU• ‘PLOTONLY’ run can loop through multiple files• Orthographic, lat/lon, polar stereographic plot projection options• Vertical cross section plots in any direction through field
•Avoid use of different models for each spatial scale;
•Multiscaling modeling;
•Numerical challenge: efficiency/precision
•Example:Global grid structure in OLAM - successor of RAMS/BRAMS
• Where are we going:
OLAM – Rainfall
OLAM – chuva acumulada kg/m^2
Satelite GOES 1999, 02 Janeiro 17:45 UTC
•OLAM - in experimental operation at CPTEC -
•Ocean model: Hycom http://oceanmodeling.rsmas.miami.edu/hycom/)
•Challenges: numerical efficiency -
•dynamical core- implicit schemes
•Sharing physics with BRAMS
•Transfer BRAMS functionalities
Conclusions:•Definition of functionalities: users (research/operational);
•No matter what model one chooses - critical to have close ties with experimental work;
•Friendly interface for users;
•Operational use (optimization function: computational cost, precision, evaluation against observations);
•Parallel efficiency is critical!
•Model validation - needs well defined metrics - appropriate for the scales;
•Team work - need to understand users needs;
•Persistence!!!
•and resources.
Recommended