CSIRO modelling and weather prediction research Jack Katzfey 9 December January 20152014
Modelling Capabilities and Research: From global to street scales 1. Global Coupled Climate modelling: ACCESS 2. Regional Modelling: Approaches
3. Regional Climate Modelling: CCAM and ACCESS 4. Downscaling Seasonal Predictions 5. Weather Prediction Research: CCAM
6. Air Quality Modelling: TAPM and CTM 7. Summary
Earth system modelling with ACCESS
Carbon cycle (ACCESS-ESM1) • • • •
Terrestrial – CABLE Ocean – Matear et al. Couple to modified ACCESS1.3 Technical coupling essentially complete • Multi-century trial simulations during 13/14
Earth system model
Atmosphere
Terrestrial
Coupler
Atmospheric chemistry
Carbon Ocean Oceanand andsea-ice sea-ice
Atmospheric chemistry • • • •
UKCA Collaboration (U. Melbourne) Couple to ACCESS-CM2 Trial simulations (coupled) during
14/15
Coupled climate and earth system modelling ACCESS1.0 and ACCESS1.3 • ACCESS1.3 uses CABLE (vn1.8) • Both among the better performing CMIP5 models • Climate change simulations to 2100 • Output freely available on Earth System grid (NCI)
Next generation model ACCESS-CM2 • Improved model physics • Higher vertical resolution • More efficient codes • Horizontal resolution – two versions • ‘Standard’ – initial for CMIP6 • ‘High’ – to support seasonal prediction
Surface air temp anomaly (°C)
CMIP5 model versions
Average skill over Globe – CMIP5 models 1000 900 800 700 600 500 400 300 200 100 0
ACCESS-1.0
ACCESS-1.3
Regional Climate Modelling Approaches Lateral boundary influence
None
High Limited area
Variable resolution
Low
Global high-resolution
High Computational expense
• Also need to consider: • • • •
Domain size and location Resolution Two-way interaction Internal variability
Downscaling methods: Grids and nesting Global Coupled Model / Analyses
Scale- selective information passed Interpolated lateral boundaries One-way information
Interaction with global circulation, teleconnections
ACCESS RCM, WRF Used for high resolution Tropical cyclone studies
Global Stretched-Grid Model (CCAM)
Scientific and technical issues • One-way nesting • No feedback to host • Necessary for ERA40 and non-native GCMs
• Two-way nesting • Feedback to host • Requires RCM to be run in-line with host • Coupling can be complex
• Variable resolution • Feedback from lower resolution regions • ENSO, SOI, SAM, etc.
The Centre for Australian Weather and
Regional Climate Modelling overview • RCMs are based around by three main components: • ‘Forcing’ regional atmospheric behaviour at ‘boundaries’ towards the host GCM
Global climate (GCM host)
• Modelling dynamical and physical processes at regional scales
Regional Climate Model
• Inclusion of surface forcings including orographic and coastal effects
Need to remove GCM biases
Nudging methods
Surface forcings
Orography and Land-use datasets, Land-surface schemes
Need to resolve features
The Centre for Australian
Cascade of uncertainty? Increased resolution Additional surface forcing Bias-correction
Uncertainty may not increase
(modified after Jones, 2000, and "cascading pyramid of uncertainties" in Schneider, 1983)
Regional Climate Modelling
Downscaling methodology • Lateral boundaries • Stretched-grid
• Domain size and placement • Stretched-grid • Scale-selective filter
• Ensembles • Downscale multiple GCMs
• Source of errors – internal and external • Bias-adjusted SSTs
• Variability – internal and external (from host model) • External same as GCMs (ENSO)
Conformal-Cubic Model
All variables are located at the centres of quadrilateral grid cells. During semi-implicit/gravitywave calculations, u and v are transformed reversibly to the indicated C-grid locations. Provides excellent dispersion properties for atmospheric waves
• • • • • •
CCAM dynamics atmospheric GCM with variable resolution (using the Schmidt transformation) 2-time level semi-Lagrangian, semi-implicit total-variation-diminishing vertical advection reversible staggering • - produces good dispersion properties a posteriori conservation of mass and moisture efficient MPI implementation
CCAM physics •
• • •
• • •
cumulus convection: - CSIRO mass-flux scheme, including downdrafts - up to 3 simultaneous plumes permitted includes advection of liquid and ice cloud-water - used to derive the interactive cloud distributions (Rotstayn 1997) stability-dependent boundary layer with non-local vertical mixing (or TKE) vegetation/canopy scheme - 6 layers for soil temperatures - 6 layers for soil moisture (Richard's equation) enhanced vertical mixing of cloudy air GFDL parameterization for long and short wave radiation Skin temperatures for SSTs enhanced for sunny, low wind speeds
Conformal Cubic Atmospheric Model (CCAM) • Developed at CSIRO for over 20 years
New features
• CCAM has run on • 25,000+ core supercomputers • Clusters • Windows laptop computers
• Chemistry transport modelling for aerosols
• Forecasts at 220 m resolution
• Dynamic ocean model • Urban model • Updated convection scheme (McGregor)
• Parallel IO and improved scaling
• Non-hydrostatic • Sophisticated (CABLE) surface scheme
• New model: flux form on gnomic grid
Bias and variance correct SST
Global Climate Model (~200 km)
Additional downscaling with spectral filter (Thatcher) Global CCAM (~50 km)
Stretched CCAM (~10 km)
MPI scaling of CCAM C192 (50CCAM km) C38435L (26 km) scaling 30O
C768 (13 km)
Simulation years per day
100
10
1
0.1
100
1000
10000
Number of cores
Fig. 5. Scaling results for single precision, atmosphere-ocean coupled CCAM simulations on IVEC’s Magnus for three different global grid resolutions. Time-steps of 900 s, 360 s and 240 s were used for the 50 km, 26 km and 13 km resolution simulations, respectively.
Summary of dynamical downscaling conducted in the PCCSP
All future time periods were simulated using the A2 emission scenario.
CRU
GCM
CCAM 60 km
CCAM 8 km
Original grid
200 km grid
TRMM
Figure 5.28: Fiji 1980-1999 annual rainfall climatology (mm/day) for TRMM satellite data (left); CRU (middle left); six global climate models model mean (middle); six CCAM 60-km multi-model mean (middle right); and three CCAM 8-km multi-model mean (right). Top row all regridded to 200 km grid, bottom row all on the original grid for each dataset.
Figure 7.17: Change in projected annual rainfall (mm/day) from additional downscaling simulations for period 2055 relative to 1990, for the A2 scenario. Note that the changes for the Zetac model are for the Jan-Feb-Mar period only . The host global climate model was the GFDL2.1 model.
Examples of climate simulations Comparison of observed trends and projected changes for 10 km ensemble mean Annual Rainfall Rainfall Observed trend Projected changes
Rainfall Change: JJA 2090-1990, RCP8.5 CCAM 50 km
Rainfall change (mm/day) 2090-1995 ~25km RCP8.5 ENSEMBLE mean
CNRM MPI
ACCESS1-0
CCSM4 GFDL
NorESM1
JJA
Quantile-quantile bias-adjustment method • 5 key variables adjusted (AWAP gridded observations) tmax, tmin, rainfall, solar radiation, pan evaporation
• Adjustments made on daily values • Each land cell adjusted independently • Adjustments made on 1 percentile bins (0.5 – 99.5) • Each season adjusted independently • Training period: 1961 – 2007 • Adjustment period: 1961 – 2100
Adjustments are small Do not change general shape of distribution Do not affect trends
Results Cressy Rainfall AWAP 1961-90 CSIRO bias-adj 1961-90 CSIRO raw 1961-90
Cressy
Frequency
oC
Cressy Tmax AWAP 1961-90 CSIRO bias-adj 1961-90 CSIRO raw 1961-90
1
Rainfall (mm)
Frequency 0
10
20
30
Temperature °C
40
10
50
Mean daily max temp for Cressy: 17.6°C (AWAP) 17.5°C (Bias adj) 17.2°C (raw) Mean annual rainfall for Cressy: 673 mm (AWAP) 683 mm (Bias adj) 1068 mm (raw)
100
Hydrology methodology Temsim CCAM
Bias-adjustment
Runoff Model
River Models
1)
Statewide Gridded Runoff
2)
Modelled 78 Catchments
3)
Modelled > 2000 subcatchments
4)
Directly used adjusted RCM outputs
How do the CCAM inputs behave in hydrological models? Black River
South Esk River 7000
1400
GCM Range
6-GCM-mean
1000
Observed
Flow (ML/d)
Flow (ML/d)
GCM Range
1200
800 600 400
6000
6-GCM-mean
5000
Observed
4000 3000 2000 1000
200
0
0
JAN FEB MAR APR MAY JUN
JAN FEB MAR APR MAY JUN
JUL AUG SEP OCT NOV DEC
Huon River
Franklin River
Flow (ML/d)
12000 10000
4000 GCM Range
3500
6-GCM-mean Observed
8000 6000 4000
Flow (ML/d)
14000
JUL AUG SEP OCT NOV DEC
3000
6-GCM-mean Observed
2500 2000 1500 1000
2000
500
0
0
JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC
GCM Range
JAN FEB MAR APR MAY JUN
JUL AUG SEP OCT NOV DEC
Bennett et al (2010)
Winter
Spring
Changes to Rivers: Seasonal Runoff Summer
No GCMs showing +ve/-ve change in runoff (Simhyd)
Autumn
Annual
Summer
Annual
Autumn
Summer
Autumn
Annual Winter
Winter
Spring
Change in runoff
Spring
Winter
1961-1990 vs 2070-2099 Summer
Spring
Summer
Autumn
Winter
Spring
Annual Autumn
Annual Percent Change in Runoff
Bennett et al (2010) Winter
-90 -60 -30 -20 -15 -10 Spring
RCM agreement
-5
5
10
15
20
30
Summer
Autumn
Winter
Spring
Annual
No. GCMs showing increase in runoff 0
1
2
3
4
5
6
6
5
4
3
2
1
0
No. GCMs showing decrease in runoff No. GCMs showing increase in runoff 0
1
2
3
4
5
6
60
90
The value of fine-scale climate modelling for hydrological impact studies Lake Crescent/Sorrell
Golden Galaxid
Clyde River Change from Baseline (1961-1990)
100%
!.
80% 60% 40% 20% 0% -20% -40% -60% -80% -100% 1990
!.
2010
2020
2030
2040
2050
2060
2070
2080
2090
2100
2040
2050
2060
2070
2080
2090
2100
Change from Baseline (1961-1990) \n
Clyde R at Hamilton
!. !.
Lower Clyde River
2000
125% 100% 75% 50% 25% 0% -25% -50% -75% -100% -125% 1990
Percent Change in Runoff
2000
2010
2020
2030
Percent Change in Runoff
-90
-60
-30 -90
-20
-60
-30
-15 -20
-10 -15
-5 -10
-5
5 5
10 10
15 15
20
20 30
30 60
60 90
90
Downscaling Seasonal Predictions • On seasonal timescales, forcing is related to SST anomalies • Downscaling approach we are developing: 1. 2. 3. 4.
Obtain SST predictions (IRI) Use current analyses Run even grid model for 1 year Downscale over region of interest
Global climate (Even grid with predicted SSTs) Nudging methods
Regional Climate Model
Surface forcings
Orography and Land-use datasets, Land-surface schemes
Weather forecasting with CCAM • Alinghi for the America’s Cup yacht race
• Rapid response systems for the Department of Defense • Australian Olympic Sailing Team • Cloud Seeding • Rural Fire Services • Renewable energy (wind farms and solar) • Operational forecasting – Predictwind.com – Melbourne 60 km Example Forecast:
8 km
1 km
220 m
Schmidt transformation can provide extremely fine resolution Grid configurations used to support Alinghi in America’s Cup C48 8 km grid over New Zealand
C48 1 km grid over New Zealand
GUI used for Microsoft Windows NWP version
Wind Validation
Validation of Rainfall extremes Extremes Validation of extremes very important. Note that the CCAM 8 km forecast extremes match observations very closely. The CCAM 1 km forecasts possibly slightly over predict extremes, though might be depend upon station locations.
Maximum daily rainfall for period 23 July 2010 to 19 July 2011. a) AWAP (25 km), b) AWAP (5 km), c) SILO (5 km), d) CCAM 60km, e) CCAM 8 km, The Centre for Australian Weather and Climate Research A partnership between CSIRO and the Bureau of Meteorology f) CCAM 1 km
www.predictwind.com (CCAM license) 520 GRIDS (360 1 KM, 160 60/8 KM) 60/8 km forecasts out to 5 days 1 km forecasts for about 1 day Two analyses (GFS and CMC) Twice per day (00 and 12 UTC) Free (until end of year) Web-based delivery Hourly images, text and grib files
The Centre for Australian
Numerical Weather Prediction • Very dependent upon initial conditions • Need to observe the atmosphere (and land and ocean) as well as possible • Need to assimilate data to form snapshot of state of atmosphere • Need to initialize model with this snapshot
• Model, using same/similar physics and dynamics as climate models, time step forward to make prediction • Accuracy dependent upon: • • • • •
Initial state Model dynamics/physics Model resolution Length of forecast (for accurate prediction of timing and amounts) Chaos of system
Locally forced versus unforced mesoscale motions • Land/sea contrasts, shape of coastline, orographic effects should all be better predicted as resolution increases, given reasonably accurate large scale patterns. • However, small-scale time and space variations over unforced regions are likely to be unpredictable. • What can models predict versus what is inherently unpredictable?
The influence of model formulation/dynamics on high resolution forecasts • The relative performance of various models used in these forecasts, all with basically the same initial state, suggests some influence of model on the quality of the forecast. • It would be interesting to investigate the error growth within the various models to determine the cause.
The value of high-resolution ensemble forecasts • Here, the ensembles were based upon different: • initial conditions • large-scale predictions and • different models.
• Usage of the global ensemble members from NCEP did not provide any useful information, since the ensemble members tended to provide significantly worse forecasts than the control. • High resolution ensemble forecasts gave indication of forecast uncertainty and range of possibilities. • At the short time scale of these forecasts (less than 18 hours), is there scope for an ensemble-based scheme which will help better capture the range of uncertainty in the 12-18 hr forecast?
Summary • Have been running 1 km CCAM forecasts since 2002, 220 m last year • Can be run on range of computers (even laptops) • High-resolution ensemble forecasts provide useful information about confidence in forecast and range of possibilities
• 1 km forecasts available at selected locations around the world • Much more validation
Modelling Urban Scales
• The Town Energy Budget (TEB) models roof, wall and road surfaces
• The scheme includes • • • • • •
Shadowing effects and reflections Conduction through walls (insulation) Internal heat source Drains/sinks to limit water and snow Heat transport with turbulent fluxes Modified roughness lengths and z0m/zoh ratios • Effect of buff bodies (i.e., not a porous canopy)
Roof
Street Canyon Building Road
Urban Heat Island – Sample from Model • In the Oklahoma city JUL2003 study we dynamically downscale to 1 km resolution
Oklahoma city, JU2003 study – Screen temperature
• Urban prognostic variables are initialised from an urban tile in the CCAM 60 km resolution simulation that has been spun-up
Urban heat island
Urban Population Pollution Exposure (urban – human scales) VECTOR WIND DIRECTION (m/s)
DAILY TOLUENE EMISSION RATES (kg per 1 x 1 km cell) (SURFACE SOURCES)
ELTH MOO
ALP FOO
RMI BXH
PAI
PTC
50
40
BRI
ANNUAL AVERAGE TOL CONCENTRATIONS (ppb) (REACTIVE)
ELTH
ELTH
MOO
ALP FOO
RMI BXH
PAI
BRI
PTC
FOO
ALP RMI
MOO BXH
PAI PTC
BRI
5 4
30
DND
DND
DND
3
20
ASP
ASP
ASP
PORT PHILLIP BAY
PORT PHILLIP BAY
2
PORT PHILLIP BAY
10 1
Emissions
Meteorology
0
POPULATION EXPOSURE (ppb-people per grid cell)
ELTH MOO
ALP FOO
RMI BXH 100000
PAI
Exposure
POPULATION DISTRIBUTION (people per grid cell)
ELTH MOO
ALP FOO
RMI BXH
PAI
80000
PTC
BRI
PTC
BRI
60000
Unreactive DND
ASP
40000
PORT PHILLIP BAY
Reactive ASP
PORT PHILLIP BAY 20000
Health Impact
0
Population
DND
0
Summary • ACCESS coupled global model • Community model developed with BoM • Submitted to CMIP5 • Developing earth system model
• CCAM: variable resolution model for regional weather and climate simulations • Completed many regional climate simulations (down to 5 km) • Used to provide specialized weather forecasts (down to 220 m) • Can run on a range of computers (supercomputers to laptops)
• CCAM-TAPM-CTM • Air quality studies and forecasts
• CCAM is also being used in: – – – – – –
India (IITM, CMMACS) Indonesia (BMKG, LAPAN) Philippines (PAGASA) South Africa (CSIR) Vietnam (IHMEN, HUS) New Zealand (NIWA)
Additional activities • Tropical cyclones in the future • Extremes, especially sub-daily rainfall for infrastructure planning • Storm surge modelling • Air quality measurements and modelling
Summary • Climate models need to take into account planetary to molecular scales
• Model needs to develop internal variability through interaction of various processes • Dynamical downscaling has similar requirements, but with added issue of using the global model information - Ensemble
• Increased resolution allows better representation of weather features and local effects • Numerical weather prediction is dependent upon the initial state of the atmosphere, as well as the need to accurately representation of weather processes - Ensemble
The Centre for Australian Weather and Climate Research A partnership between CSIRO and the Bureau of Meteorology
Thank you Jack Katzfey CSIRO Oceans and Atmosphere Flagship Phone: +61 3 9239 4562 Email:
[email protected]