RegClim: Local and regional climate scenarios for Norway

Regional Climate Development under
Global Warming
(www.regclim.met.no)
Norges Forskningsråd:
Seminar om bruk av klimascenarier i virkningsstudier, Oslo, 16.04. 2004
RegClim: Present results and future plans relevant for climate
impact studies
E.J.Førland, R.E.Benestad, I.Hanssen-Bauer, T.E.Skaugen, J.E. Haugen, D. Bjørge, M.
Ødegård, E.Støren, A.Sorteberg, B.Ådlandsvik
• Uncertainty and confidence in present RegClim scenarios
• Downscaling scenarios for impact studies: Tools and
examples
• Tailoring scenarios for specific impact assessments:
 What RegClim can and cannot produce
RegClim’s Overall Aim Phase III
• to produce scenarios for regional climate change
– suitable for impact assessments in Northern Europe,
bordering sea areas and major parts of the Arctic (our
region), given a global climate change;
• and to quantify uncertainties
– due to choice of methods, global scenarios, and due to
poorly understood processes influencing our region's
climate, in particular:
• those causing the warm and ice-free Nordic Seas,
• and the effects of aerosols.
Keywords: Risks and uncertainties
Regional Climate Development under
Global Warming regclim.met.no
RegClim: 5 Principal Modules (PM)
Regional risk
and uncertainty
PM1:
Impacts
Atmospheric
interpretation
for regional climate.
Eirik Førland, met.no.
Global-scale response
PM3:
Global and regional
significance of the
Nordic Seas.
Nils Gunnar Kvamstø,
UiB
PM2:
Regional interpretation
for
oceanic and Arctic
climate.
Bjørn Ådlandsvik, IMR.
PM4:
Climate response of
regional contaminants.
Jon Egill Kristjansson,
UiO.
Basic theory
PM5:
Optimal forcing
structures for
atmospheric flows &
regional climate
predictability.
Trond Iversen, UiO
Regional Climate Development under
Global Warming regclim.met.no
Regional Climate Development
under Global Warming
www.nilu.no / regclim
WARNING !
For interpretation of climate scenarios (e.g. for impact
assessments) it is important to keep in mind that:
•
A climate scenario is an internal consistent realisation of
the future climate development under certain
assumptions
•
Climate model simulations with slightly different initial
conditions may exhibit different characteristics in each
evolution (Large variability and some uncertainties…!)
•
A climate scenario is not a forecast !
Meteorologisk Institutt met.no
Large uncertainty North of 60oN
J. Räisänen:
Increased Temperature by 2080 rel. to Global average for 19 CMIP2 runs
Red curve is the average over all 19.
Uncertainties in future climate projections
•
•
I. Global System Unpredictability
–
–
Natural forcing: Solar radiation, volcanoes
Anthropogenic release of gases and particles
–
Changes in land-use
II. Climate System Unpredictability:
–
•
(“climate forcings”):
Internal variations in the climate system (“Natural variability”)
III. Model Deficiencies:
–
Imperfect knowledge about forcings and processes
–
Imperfect physical and numerical treatment of processes
–
Poor resolution in the global models
–
Imperfect downscaling procedures
I. Global System Unpredictability
II. Climate System Unpredictability / III. Model Deficiencies:
RegClim Regional Climate Model (HIRHAM)
Precipitation response winter DJF (mm/day)
MPI (GSDIO, 1980-1999 +50)
Hadley (A2, 1961-90 + 110)
Is this due to model uncertainty only?
Can sample uncertainty be an explanation too?
Uncertainties II & III ctd.:
PM3: Bergen Climate Model Control Run 300 yrs: Atlantic Meridional Overturning
(AMO). CMIP2: 4 runs starting from different AMO-strength:
Notice: E77: AMO=19Sv ; E78: AMO= 16 Sv.
AMOMEAN
= 17 Sv
Nils Gunnar Kvamstø and Asgeir Sorteberg, Univ. of Bergen
Uncertainties II & III ctd.:
E77
Start from
strong AMO
”MPI-type”
Winter
delta Precip.
mm/day
At 2 x CO2
E78
Start from
weak AMO
”Hadley-type”
ANSWER: Yes it may be due to sample uncertainty.
Norwegian Meteorological
Norwegian
Institute met.no
Downscaling in RegClim
Why?
• Poor spatial resolution in
global models
How?
• Dynamical downscaling (regional modelling).
• Empirical downscaling (statistical
downscaling).
Spatial resolution
1x1
km2
55x55 km2
Downscaling techniques applied by RegClim
(“et al.”)
• Dynamical downscaling (Regional Climate Models,
RCM)
– Grid points (55x55km, will be refined to 22x22 km)
– ”Δ-Change”
– Interpolation to specific sites (different elevation,
distance to coast, local topography, …)
– Interpolation + adjustment to observed climatology
– Empirical downscaling based on RCM-values
• Empirical downscaling (Statistical downscaling)
– Linear techniques (regression [simple + multiple, CCA,
SVD, etc])
– Non-linear techniques (Analogues [Weather types],
Weather Generator, Neural networks, etc)
Dynamical downscaling
–
–
–
–
(RegClim-contact: [email protected])
A “Regional Climate Model (RCM)” is run with high spatial
resolution for a limited area
Results from a global model are used as input at the
boarders of the RCM
The RCM-simulations are performed for present
(“control”) and future (“scenario”) climate
A number of climate elements are calculated for every 6
hours, e.g.:
•
•
•
•
•
•
•
Sea level air pressure
Temperature (2 m): mean, maximum, minimum
Wind-speed and –direction (10 m)
Precipitation-amount (rain & snow)
Snow depth
Evaporation
Wind and geopotential height for the 500 hPa-level
Meteorologisk Institutt met.no
Scenarios from dynamical downscaling
• ”MPI”: ECHAM4/OPYC3 GSDIO,
IS92a
– 1980-99 & 2030-49
– 3 different domains
• ”Hadley”: HadAM3h, A2
– 1961-90 & 2071-2100
– Domain 3
• Several meteorological
elements are available with a
time resolution of 6 hours in a
55x55 km grid and for 19 levels
Projected change, winter temperature
Small differences between simulations
MPI (IS92a)
Hadley(A2)
Projected change, winter precipitation
Substantial differences between simulations
MPI IS92a
Hadley A2
Jan Erik Haugen, PM1:
Combined statistics of HIRHAM downscaled
scenarios, 55x55 km
•
ECHAM4 IS92a, 1980-99 and 2030-49
1 2x20 years, 50 year response
•
HadAM3H A2, 1961-90 and 2071-2100
1 2x30 years, 110 year response
Scaling procedure
The Hadley response over 110 year was scaled to the 50 year
MPI response period from the global temperature of the Hadley
simulation -> Factor 0.32 for Hadley data
(Christensen et al, 2001, Geophys. Res. Letters, 28,1003-6)
The two (scaled) scenarios were treated as equally valid
realizations i.e. it was assumed that the difference is due to
natural climate variability
Precipitation response winter DJF
MPI
Hadley
Combined
(unscaled)
(after scaling)
Monthly projections from empirical
downscaling by regression models
• MPI: ECHAM4/OPYC3 GSDIO, IS92a
– Temperature and precipitation for ca 50
Norwegian stations are downscaled for the
period 1860-2050
Empirically downscaled temperature scenario for Longyearbyen
LONGYEARBYEN - ANNUAL MEAN TEMPERATURE
2
0
Temperature, deg C
-2
-4
-6
-8
-10
-12
OBS
-14
1900
1925
1950
1975
2000
2025
Winter temperature,
Longyearbyen
100
90
80
obs1912-49
obs1950-99
mod1900-49
mob1950-99
mod2000-49
70
60
50
40
30
20
10
0
-26
-24
-22
-20
-18
-16
-14
Temperature, deg C
-12
-10
-8
-6
-4
MOD
2050
Monthly projections from empirical
downscaling by regression models
• MPI: ECHAM4/OPYC3 GSDIO, IS92a
– Temperature and precipitation for ca 50 Norwegian stations
are downscaled for the period 1860-2050
• ”Multimodel-experiments”:
– Several global models are downscaled for a number of
stations in Northern Europe
– The downscalings are based on different choices of domains
and predictors
• Selected scenarios are available through NOSerC
(http://noserc.met.no/effect/)
Contact person: Egil Støren
Uncertainty in global models & downscaling
Projected warming in Oslo in January,
results from 48 scenarios (17 global models):
Robust signal:
Warming
Uncertainty in global models & downscaling
Empirical downscaling of various GCMs
Larger spread for precipitation than for temperature
– Uncertainty related to downscaling techniques
Projected change in annual mean temperature during
50 years: Results from dynamical (DD) and empirical
downscaling (ED) (MPI ECHAM4/OPYC3 GSDIO, IS92a)
Rather similar values
Projected change in annual mean temperature
2.5
DD
Deg C
2.0
ED
1.5
1.0
0.5
0.0
R1
R2
R3
R4
R5
R6
– Uncertainty related to downscaling techniques
Projected change in annual precipitation during 50
years: Results from dynamical (DD) and empirical
downscaling (ED) (MPI ECHAM4/OPYC3 GSDIO, IS92a)
 Some deviations.
Projected change in annual precipitation
20
% of 1980-1999 mean
18
16
DD
ED
14
12
10
8
6
4
2
0
RR1 RR2 RR3 RR4 RR5 RR6 RR7 RR8 RR9 RR10 RR11 RR12 RR13
RegClim PM1: Present activities/future plans
Quantifying sources of risks and uncertainties by:
• Downscaling simulations with the same emission scenario
from different global models (uncertainty in global models);
• Increasing available downscalings through Nordic/European
collaboration (uncertainty in regional models);
• Empirical vs dynamical downscaling (methodological
uncertainty);
• mapping regional natural variability by:
– Downscaling ”ensemble runs” from the Bergen Climate Model
(PM3) which is focussing the Atlantic Ocean currents;
– Downscaling global scenarios from the Oslo GCM (PM4) which
is focussing the influence of aerosols;
Length of the growing season in south-eastern part of Norway
a)
1961-1990
b) (2021-2050)-(1961-1990)
Meteorologisk Institutt met.no
Average change in the snow storage
by 1.April
• The snow storage is
expected to increase at
altitudes above 800 m in
Eastern Norway as a result
of increasing winter
precipitation
• Large spring floods are
therefore possible even in
a warmer climate
(NVE & met.no)
RegClim: Downscaling of daily values
Main elements: Temperature and precipitation
Simulation
Emission
Domain
Method
Period
-
D1,D2,D3
D:RCM
1979-1993
F
ECHAM4/OPYC3
IS92a
D1
D:RCM
1980-1999, 2030-2049
F
ECHAM4/OPYC3
IS92a
D2
D:RCM
1980-2049
F
ECHAM4/OPYC3
IS92a
D3
D:RCM
1980-1999, 2030-2049
R
ECHAM4/OPYC3 T106
B2
D2
D:RCM
1961-1990, 2071-2100
P
HadAm3H
A2
D3
D:RCM
1961-1990, 2071-2100
R
HadAm3H
B2
D3
D:RCM
1961-1990, 2071-2100
P
ECHAM4/OPYC3
IS92a
D2
E:Analog
1980-2049
F
ECHAM4/OPYC3
IS92a
D2
E: Linear
1980-2000,2030-2049
F
A2
D3
E:Analog
1961-1990, 2071-2100
P
ERA-15 (1979-93)
HadAm3H
State
Domains: D1: Large, D2: Medium, D3: Small
Method: Dynamical/RCM, Empirical: Linear regression & Analogs
State: F: Finalised, R: Downscaled, but not finally quality-checked, P: Planned
downscaling
RegClim PM2: Info on marine impact studies
RegClim contact-person: [email protected])
• To study effects on marine ecosystems, it is not sufficient
to know the state on wind, air temperature and
precipitation. Coupled global atmosphere-ocean models
may be used, but the resolution is too coarse to resolve
the topography on the continental shelves
• In RegClim PM2 regional marine scenarios will be
established, with emphasis on currents, salinity and
temperature from the sea surface to the bottom in the
ocean areas around Norway
• This is a new activity, and presently few results are
finalized. Available at the moment is a scenario with
ocean level (storm surge) and waves
Nature, January 11, 2004:
Schär et al.: The role of increasing variability in European summer heatwaves
Various users may have different needs
for climate data for impact studies
• High resolution in time and space
• Representative for local climate conditions
Cumulative frequency distribution, annual temperature, Oslo
100
90
80
70
Frequency, %
• Climatology (Mean values, extremes,
frequencies, etc.)
• Internal consistency (spatial, temporal [dayto-day] and between elements)
• Optimal historical time series (observations)
to ”calibrate” empirical models
60
obs1900-49
50
obs1950-99
40
mod1900-49
mob1950-99
30
mod2000-49
20
10
0
2
2.5
3
3.5
4
4.5
5
5.5
6
6.5
Temperature, deg C
7
7.5
8
8.5
9
9.5
10
Benefits and drawbacks of different downscaling methods
High resolution
in
Repr.
local
Internal
consistency
Internal
consistency
Space
Time
conditions
Statistics
day-to-day
betw.
elements
Multiple
scenarios
Grid points (RCM)
N
Y
N
Y
Y
Y
N
D-change
N
Y
N
Y
?
?
N
Interpolation
N
Y
N
Y
Y
Y
N
Interpol+adjust
?
Y
Y
Y
Y
Y
N
Simple regr.
N
N
Y
Y
N
N
Y
Multiple regr.
?
N
Y
Y
N
?
Y
Analogues
Y
N
Y
Y
N
Y
Y
DYNAMICAL
Emp.downsc.
EMPIRICAL
Weather Gen.
YES
NO
MAYBE ?
RegClim Phase III: Present activities/future plans, PM1
Produce the type of scenario data needed for impact
assessments:
• Implementing RCMs with higher spatial resolution;
• Empirical downscaling of daily values;
• Improve methods for estimating changes in “climatic risks”, i.e.
changes in frequency distributions and extremes of
temperature, precipitation, wind, waves, etc;
• Improve scenarios for changes in snow conditions;
• Develop user-friendly tools for empirical downscaling;
• Make results available on the web: (http://noserc.no/effect);
•
NB! Tailoring of climate scenarios for specific impact studies is
NOT a part of the planned RegClim activities
 A dialogue between users &RegClim is important to ensure:
- That appropriate data for impact assessments are focussed
- That interpretation of the results is optimal
•
Overview: Main features of downscaled scenarios
for Norway (MPI IS92a & HadCM A2)
•
Temperature: MPI og Hadley in average give similar warming per decade.
The scenarios indicate larger warming during winter than summer, larger
warming in inland than at the coast, and larger in the northern than
southern regions
•
Rare extremes (compared to present climate) probably will occur more
frequent for high temperatures and less frequent for low temperatures
Precipitation: Locally there are rather large differences between the MPI
and Hadley simulations. This may be caused by natural variability
High intensity events probably will occur more frequent in the future
•
•
•
•
Snow: Both models indicate less snow in the lowland parts, particularly in
the coastal regions. In the inland areas and particularly in the highmountains the models indicate risk of increasing snow amounts during
winter
Atmospheric circulation / Storm tracks: Substantial differences between
MPI and Hadley in projected changes over Norway. This implies differences
for e.g. local scenarios of precipitation and wind