Transcript Downscaling methodology

Downscaling Methodology
Dr Jack Katzfey
CLIMATE ADAPTATION FLAGSHIP
December 2012
Process of understanding:
• Observations
• What time and space scales are needed?
• Understanding
• Explaining the process of what we see
• Modelling
• What complexity is needed for what scale?
• Prediction
• Forecasts versus projections
• Adaptation
• What will help reduce affect of climate change?
SPATIAL SCALES – WORLDCLIM example
GlobalRegional
Scaled
(‘Delta method’*)
Pattern scaling
(Marksim Weather Generator)
GCM
200-300 km
Statistical downscaling
Dynamical downscaling
(PRECIS)
✗
✔
RegionalSite
✗
✗
✔
✗
✔
✗
*The method makes the following gross assumptions:
1. Changes in climates vary only over large distances
(i.e. as large as GCM side cell size)…..
1 km
climate
surfaces
Downscaling aims
• To provide more detailed (and hopefully more accurate)
information on current and future regional climate through higher
resolution simulations with better resolved physical processes and
surface inputs. The more detailed orography and land use
information input to the regional model provides more information
than coarser resolution analyses and more spatially consistent
information than gridded observations. In addition, the finer
resolution of the regional model should more realistically
represent atmospheric phenomena and dynamics.
Downscaling aims
• To simulate the current climate accurately. Future climate change
will then be superimposed on a more realistic current climate,
which in turn should give more confidence in climate change
projections.
• To produce a range of downscaled climate change signals in order
to capture the range of possible future climate projections,
consistent with the range seen in the large-scale GCMs on which
the RCMs are based. This is the approach taken in the IPCC
projections of climate change, where it is acknowledged that there
are biases and uncertainty in model projections, so that the
consideration of the output of as many models as feasible, in what
is known as ensemble predictions, is the most reasonable way to
deal with uncertainty.
What is Regional Climate Modelling?
• RCMs are based around three main
components:
• Nudging regional atmospheric behaviour
at ‘boundaries’ towards a host GCM
• Modelling dynamical and physical
processes at regional scales
Global climate
(GCM host)
Regional Climate
Model
• Inclusion of surface forcings, including
orographic and coastal effects
Surface forcings
Regional Climate Modelling
Nudging/forcing/LBC
Dynamical and
physical
parameterisations
Domain/Resolution
Land-surface
schemes/
specification
Motivation for bias correction
Bias correction
GCM (~200 km)
RCM (~60km)
RCM (~8 km)
Bias correction
What is Regional Climate Modelling?
• RCMs are based around three main
components:
• Nudging regional atmospheric behaviour
at ‘boundaries’ towards a host GCM
• Modelling dynamical and physical
processes at regional scales
Global climate
(GCM host)
Regional Climate
Model
• Inclusion of surface forcings, including
orographic and coastal effects
Surface forcings
Regional Climate Modelling
Nudging/forcing/LBC
Dynamical and
physical
parameterisations
Domain/Resolution
Land-surface
schemes/
specification
What is Regional Climate Modelling?
• RCMs are based around three main
components:
• Nudging regional atmospheric behaviour
at ‘boundaries’ towards a host GCM
• Modelling dynamical and physical
processes at regional scales
Global climate
(GCM host)
Regional Climate
Model
• Inclusion of surface forcings, including
orographic and coastal effects
Surface forcings
Regional Climate Modelling
Nudging/forcing/LBC
Dynamical and
physical
parameterisations
Domain/Resolution
Land-surface
schemes/
specification
Regional Climate Modelling Approaches
Lateral boundary influence
None
High
Variable
resolution
Limited area
High
Low
Computational expense
• Also need to consider:
•
•
•
•
Global
high-resolution
Domain size
Resolution
Two-way interaction
Internal variability
Regional Climate Modelling
Comparison of GCM, LAM (ACCESS
RCM) and SGRCM (CCAM) grids
Scale- selective filter
(frequency domain)
Interpolated lateral
boundaries
GCM
One-way
nesting
Limited Area Model (LAM)
Global Stretched Grid Model (SGM)
Processes in Climate Models
Mike Manton APN Symposium 2004
The Centre for Australian Weather and Climate Research
A partnership between CSIRO and the Bureau of Meteorology
Limited Area Models - BC
• Large amounts of data are needed in order to specify the vertical
profile of data at all lateral boundary points frequently enough
(typically 6 hourly) in order to capture the atmospheric flow
realistically, including possible diurnal effects.
• There may be undesirable effects in the LAM when the host data is
interpolated to the finer LAM grid.
• Potentially, large gradients can occur in the boundary region as the
internal model (LAM) drifts to a different climate then the host
model (GCM).
• Differing resolution, physics and dynamics can lead to differences
in evolution of weather systems within the LAM, so great care
must be used in formulating the lateral boundary scheme.
Limited Area Models - BC
• The location of lateral boundaries (i.e., they should ideally not be
in regions of significant orography)
• The amount of change in resolution between host and LAM (one
does not want to introduce damping or inconsistent data when
the host model data needs to be interpolated in space and in time)
• How much two-way interaction is allowed (i.e., interaction of the
regional simulation with the larger scale), to ensure that larger
scale variability from phenomena such as El Niño-Southern
Oscillation (ENSO) is transferred into the domain.
Strengths of LAMs:
• There is less computational cost than for SGM, since you only
compute for the area of interest.
• Only regional surface input datasets are required.
• The model can be more optimally configured for the model’s given
resolution, which is not possible if there are a variety of
resolutions, as for stretched-grid RCMs.
• Larger user communities for technical support.
Weaknesses of LAMs:
• Treatment of the lateral boundary conditions is difficult. There are
problems of passing information from the host model through the
lateral boundaries and potential incompatibilities of internal
systems passing into the boundary zone.
• There is potential incompatibility of the regional model with the
boundary data due to different scales.
• A large amount of atmospheric data is required to provide
sufficient information at the lateral boundaries to drive the model.
• Internal model climatology may be different than boundary data,
leading to boundary problems.
Strengths of Stretched-Grid RCMs:
• There are no lateral boundaries.
• There is a potential to correct some of the biases in the input data
through bias correction techniques.
• There is a potential for two-way interaction between the high
resolution area and outside regions. (Teleconnections?)
• Potentially less data is required to run than for a LAM if only
surface input data from host model is required. (Unless large-scale
nudging is used, than more data is required than for RCMs)
Weaknesses of Stretched Grid RCMs
• The model needs to be configured to run correctly at a range of
horizontal resolutions.
• Global surface datasets are required to run the model.
• Global datasets of other atmospheric variables are needed if
atmospheric nudging is required. Because they are global, these
datasets will be larger than the regional ones required for LAMs.
• Greater need for conservation of atmospheric properties is
required.
• Potential weakness is if run with only lower boundary condition,
model may generate different atmospheric response –
interpretation?
Regional Climate Modelling overview
• A RCM is not simply a ‘long’ NWP run
• `Boundary’ conditions are particularly important for
determining how well a regional model captures all
drivers that influence a region’s climate (including all
forcings from the host GCM)
• A RCM will tend to spin-up its own local climate
behaviour based on reconciling the atmospheric and
surface forcings in a way that is consistent with the
model’s dynamical and physical formulation (but what
about model errors?)
• Seamless prediction? (forecast errors=climate errors)
Regional Climate Modelling
Multiple downscaling to higher resolution
Simulated annual rainfall for Tasmania at different resolutions
Global Model
Regional Climate Modelling
CCAM 60 km
CCAM 14 km
Bias-correc.
Spectral
forcing
Increased
resolution
Increased
resolution
Climate Futures for Tasmania project
Ensembles
Change in annual rainfall 1961:1990 to 2070:2099
14 km results
Although mean changes,
pattern fairly consistent
Regional Climate Modelling
Climate Futures for Tasmania project
21
Validation
of
DDS
When forced by GCM (via
CCAM), slight degradation
(CCiP, 2011)
Annual rainfall change (mm/d)
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 JanFeb-Mar period only . The host global
climate model was the GFDL2.1
model.
(CCiP, 2011)
Future directions – Urban modeling
Urban model within RCM
From H. Schluenzen, Univ. of Hamburg
Regional Climate Modelling
Methodology – urban climate
The aTEB scheme includes the
following additional features:
• Alternative in-canyon aerodynamical
resistance scheme including
recirculating and venting regions
(based Harman et al 2004)
• Two canyon walls rotated through
180o instead of a single TEB wall
rotated through 360o
roof
wall
• Big-leaf model for in-canyon
vegetation (similar to Sang-Hyun and
Soon-Ung 2008, but using Kowalczyk
et al 1994)
wall
• Modified in-canyon reflections for
radiation to account for the extra wall
AC
• Simple AC heat flux into canyon (see
also Ohashi et al, 2007)
traffic
road
Schematic representation of the aerodynamic resistances.
Note snow and water have been omitted in this diagram
The Centre for Australian Weather and Climate Research
A partnership between CSIRO and the Bureau of Meteorology
Simple forecast experiment of impact of urban model
of forecast due to resolution
• CCAM intialised with NCEP fnl analysis at 00 UTC 5 January 2003
• Used aTEB (Thatcher and Hurley, 2012) with default Melbourne
urban settings (Coutts et al, 2007)
• Forecasts for 3 days at:
• 60 km over all of Australia
• 8 km over Victoria with spectral forcing from 60 km forecast
• 1 km over Melbourne with spectral forcing from 8 km forecast
• Two experiments using CABLE land-surface scheme:
• One without urban scheme (nu)
• One with Urban tiles (u)
T2m (°C) time series in city (Day 2-3)
42
• Black with
urban
• Red w/o urban
60 km
uf=.25
2
18
8 km
uf=.97
1 km 6
uf=1
• Difference U-NU
• Note greater affect or
urban at higher
resolution
• Melb. R.O.:
• Tmax: 34.9
• Tmin: 16.6
• Tmax: 40.2
Nocturnal temperature diff. due to urban
• Left: T2m (urban)T2m(non-urban) avg.
• 11pm-5am (12-18 UTC)
• Right: Urban fraction
• Top: 60km,
• Middle: 8 km and
• Bottom: 1 km forecasts
Winds(10m) 12pm-6am
Wind differences (u-nu)
(scale 2m/s)
Shading surface height (m)
Winds (scale 10m/s)
 Note land breese and affect of urban on winds
Future Work
• Linking to observed trends
• Relationship to statistical downscaling
• Linking into Climate Futures tools
• GCM selections (regional SST changes?)
• More analysis of storm tracks and teleconnections
• New version of CCAM and ACCESS RCM
• Coastal Effects
• Hydrology, waves, salinity
• Development of a Regional Earth System Model
• Air quality
• Renewables
• Integrated assessment
Regional Climate Modelling
Summary
• The effect of resolution on the urban environment
• Partly due to urban fraction at the coarse resolution?
• Note in regional climate mode, we cannot afford to run at very highresolutions for long simulations
• CCAM capability of forecasts (as shown here)
• CCAM capability as a regional climate model (as
shown by Marcus Thatcher)
• Ensembles
• Bias-correction
• Many experiments can be run…
• Cost effective – Not that expensive!
Adding Value Through Downscaling
• Finer resolution
• More realistic surface forcing
• Higher resolution surface land-use and orography
• Multiply downscaling to finer resolutions
• Reducing GCM errors
• Bias-adjusted SSTs
• Ensembles
• Downscale multiple GCMs
• Multiple RCMs
• Multiple parameterizations/parameter ensemble?
Regional Climate Modelling
Why use regional climate models?
• Can provide a spatially and temporally consistent
dataset
• Can provide finer resolution datasets
• Can provide physically-based effects caused by local
forcings (especially orography)
• Small, statistically based corrections can be applied if
needed
• Caution:
• Technique can influence results
• Needs to be validated
Regional Climate Modelling
Applications
• Current climate
• Potential to assess climate in regions without observations
(higher resolution than traditional ‘reanalyses’
– But not site specific (like wind farm siting)
• Assessment of ability of models to simulate current climate
Regional Climate Modelling
Downscaling from Reanalyses
• Wind parameters derived from
REMO data are in agreement
with observations, and on
average, they describe the wind
magnitude slightly better than
the NCEP/NCAR re-analysis
data.
Larsén et al., Wind Energy 2010; 13:279–296
Regional Climate Modelling
Multi-year forecasts or downscaling from analyses:
Provides statistics of winds, even when no observations are available
10.0
2.0
1.0
0.5
0.2
Frequency-weighted variance spectrum
0.1
• Modelling the wind climatology
• Modelling the variance spectra
• Modelling spatial correlations between sites
•The results are then used to anticipate the
impact of wind farms on the electricity
market (i.e., highly correlated output
between wind farms can lead to network
instability)
0.005
0.010
0.020
0.050
0.100
0.200
0.500
Frequency (cycles/hour)
Site: WLN
0.001
0.005
Probability density
0.500
Tower (8km cell)
CCAM 8 km
0.050
•CCAM is being used to study the aggregate
behaviour of multiple wind farms, including:
Tower
Tower (8km cell)
CCAM 8 km
5.0
Wind farm research
0
5
10
15
Speed (m/s)
Site: WLN
Regional Climate Modelling
20
25
30
Results
from
WERU
Applications
• Future climate projections
Regional Climate Modelling
Cascade of uncertainty?
Need for ensembles
(modified after
Jones, 2000, and
"cascading
pyramid of
uncertainties" in
Schneider, 1983)
Regional Climate Modelling
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
Climate change vs model uncertainty
• Future climate change signal composed of:
• Climate change sensitivity/signal
• Model errors (especially current climate)
• Inaccurate responses of models to climate change forcing
• Increasing confidence of regional climate change projections
• Higher resolution
• More realistic surface forcing
• Possibly reduced errors in current climate
• Ensembles
Regional Climate Modelling
Large-scale SST bias-correction
Surface temperature average
115 E to 155 E, 40 S to 10 S
3 year running average
Model uncertainty
plus change (3.7°C)
• In addition to fixing biases,
allows simulation to have
more realistic weather
systems and how they may
change in response to
climate change
Model uncertainty/error (2.3°C)
Example of SST bias in a GCM
Spread of
change signal
(1.4°C)
Same mean
Regional Climate Modelling
Bias adjustment of sea surface temperatures
• Sea surface temperatures is main influence on
climate (ENSO, climate change)
• Dommenget, Dietmar, 2009: The Ocean’s Role in
Continental Climate Variability and Change. J.
Climate, 22, 4939–4952
• Can improve the representation of the current
climate by fixing some of the biases
• Ensemble using only one downscale model
does not decrease spread of climate change
signal
Regional Climate Modelling
43
Modes of running CCAM
• Numerical Weather Prediction
• Seasonal Prediction
• Regional Climate Prediction
Numerical Weather Prediction
• Model set-up as usual, but potentially at higher
resolution
• Multiple downscaling since runs are for shorter time
scales (1 day to couple weeks)
• Initial condition very important
• Atmospheric fields
• Surface fields, such as soil moisture and temperatures
• Time varying ocean temperatures and GH gases not as important
Seasonal Prediction
• Model set-up as usual, but potentially at `modest’
resolution
• Multiple downscaling since runs are for shorter time
scales (from months to 1 year?)
• Initial condition important (but maybe less so than for
NWP)
• Atmospheric fields
• Surface fields, such as soil moisture and temperatures
• Time varying ocean temperatures are very important
• But GH gases may be fixed
Regional Climate Prediction
• Model set-up as usual, but potentially at `lowest’
resolution
• Multiple downscaling (but costly)
• Initial condition not as important
• Atmospheric fields
• Surface fields, such as soil moisture and temperatures
• Time varying ocean temperatures are very important
• GH gases and aerosols need to vary over time