n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Verification of temperature and precipitation trends in climate models.
Geert Jan van Oldenborgh, KNMI many co-authors
• Introduction
• Uninitialised climate simulations – Observed temperature trends
– Climate model temperature trends
– Statistical analysis of the discrepancies – Physical causes of the discrepancies
• Initialised climate simulations – Global temperature
– Temperature fields – AMO index
– Precipitation fields
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Oldenborgh, G.J. van, S.S. Drijfhout, A. van Ulden, R. Haarsma, A.
Sterl, C. Severijns, W. Hazeleger and H. Dijkstra, Western Europe is warming much faster than expected. Climate of the Past, 2009, 5, 1-12.
Oldenborgh, G.J. van, F. Doblas-Reyes, W. Hazeleger and B.
Wouters, Skill in the trend and internal variability in a multi-model decadal prediction ensemble, in preparation
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Temperatures at De Bilt 1706-2009
-4.00 -3.00 -2.00 -1.00 0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
1700 1750 1800 1850 1900 1950 2000 2050
Dec-Feb
5.00 6.00 7.00 8.00 9.00 10.00 11.00 12.00
1700 1750 1800 1850 1900 1950 2000 2050
Mar-May
13.00 14.00 15.00 16.00 17.00 18.00 19.00
1700 1750 1800 1850 1900 1950 2000 2050
Jun-Aug
7.00 8.00 9.00 10.00 11.00 12.00 13.00 14.00
1700 1750 1800 1850 1900 1950 2000 2050
Sep-Nov
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Central Netherlands Temperature 1906-2009
-4.00 -3.00 -2.00 -1.00 0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
1900 1920 1940 1960 1980 2000 2020
Dec-Feb
6.50 7.00 7.50 8.00 8.50 9.00 9.50 10.00 10.50 11.00 11.50 12.00
1900 1920 1940 1960 1980 2000 2020
Mar-May
14.00 14.50 15.00 15.50 16.00 16.50 17.00 17.50 18.00 18.50 19.00
1900 1920 1940 1960 1980 2000 2020
Jun-Aug
7.00 8.00 9.00 10.00 11.00 12.00 13.00 14.00
1900 1920 1940 1960 1980 2000 2020
Sep-Nov
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Comparison with climate model (17 × ECHAM5)
-2 -1 0 1 2 3
1950 1960 1970 1980 1990 2000 2010 2020 2030 Essence ensemble
world averaged observations
-2 -1 0 1 2 3
1950 1960 1970 1980 1990 2000 2010 2020 2030 Essence ensemble
CNT observations
Global mean temperature Central Netherlands Temperature
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Definition of trend
• The difference of averaged periods is very noisy due to the step functions at the edges of the intervals. 1997 shifting in and out of the averages makes a big difference.
• A linear trend is better, but still depends on the starting date.
Starting in 1963 gives a different result from 1940 or 1975. Also, the trend is not linear.
• Therefore the trend is defined by the regression against global mean temperature. This method gives rise to smaller residuals.
T(x, y, t) = A(x, y)Tglobal(3) (t) + (x, y, t) (1) (A 3-yr running mean is applied to Tglobal (t) to filter out ENSO effects.)
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Definition of trend
-0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5
1950 1960 1970 1980 1990 2000 2010
Ta [Celsius]
HadCRUT3 Tglobal
14 14.1 14.2 14.3 14.4 14.5 14.6 14.7 14.8 14.9
1950 1960 1970 1980 1990 2000 2010
temp2 [Celsius]
Essence Tglobal
Fingerprint function.
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Observations and models
Observations:
• De Bilt homogenised temperature, Central Netherlands Temperature, CET
• CRUTEM3 and HadSST2, weighted according to fraction land/sea
GCM ensembles:
• ESSENCE: 17-member ECHAM5/MPI-OM 1950-2100 (20C3M, SRES A1B) T63,
• 5 models from CMIP3 with the most realistic circulation over Europe (ECHAM5/MPI-OM, GFDL CM2.1, MIROC 3.2 T106, CCCMA CGCM 3.1 T63, HadGEM1), see van Ulden and van Oldenborgh ACP 2006,
• full CMIP3 ensemble, 22 models (left out GISS-EH),
• UKMO QUMP perturbed physics ensemble, 17 members, 1850-2100, 20C3M/SRES A1B.
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Global mean temperature
-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
1880 1900 1920 1940 1960 1980 2000 no volcanoes
with volcanoes GISS Ts obs HadCRUT3 obs Mauna Loa CO2
• Without volcanoes, Tglobal follows the CO2 concentration
• With volcanoes the models track Tglobal quite well.
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Forecasting global mean temperature
290 300 310 320 330 340 350 360 370 380 390
1880 1900 1920 1940 1960 1980 2000 2020
CO2 [ppm]
-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8
290 300 310 320 330 340 350 360 370 380 390
annual Tglobal anomalies
annual CO2 concnration [ppm]
NCDC global temperature vs CO2 1880:2009
1.1880 1881 1882 1883 1884 18851886 1887 1888 1889
1890 1891 1892 1893 1894 1895 18961897
1898 19001899 1901 1902
1903 1904
1905 1906
1907190819091910 1911 19121913 1914 1915
1916 1917 191819191920 1921 192219231924
1925 1926 19271928
1929 1930 1931 1932
1933 193419351936 1937193819391940 1941 19421943 1944
1945
194619471948 1949 1950 1951 1952 1953
19541955 1956 19571959
1960 196119621963
1964 1965 196619671968
1969 1970
1971 1972
1973
1974 1975
1976 1977
1978 1979
1980 1981
1982 1983
19841985 1986
19871988 1989
1990 1991
19921993 1994
1995
1996 1997
1998
19992000 2001
20022003 2004
2005 20062007
2008 2009
r = 0.895
• CO2 concentrations increase by 1.94(2) ppm/yr since the collapse of communism,
• Global mean temperature rises with 0.0086(4) K/ppm over 1880-2009,
• Hence we expect T to rise with 0.017(1) K/yr.
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Local vs global warming
CRUTEM3+HadSST2 observations ESSENCE, 17 runs with ECHAM5/MPI-OM
Main features are very similar, how about the details?
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Local vs global warming
CRUTEM3+HadSST2 observations GFDL CM 2.1
Main features are very similar, how about the details?
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Local vs global warming
CRUTEM3+HadSST2 observations CCCMA CGCM 3.1 T63
Main features are very similar, how about the details?
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Local vs global warming
CRUTEM3+HadSST2 observations MIROC 3.2 T106
Main features are very similar, how about the details?
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Local vs global warming
CRUTEM3+HadSST2 observations UKMO HadGEM1
Main features are very similar, how about the details?
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Significances of trend differences
There are two fundamentally different questions one can ask
Is the observed trend withn the PDF defined by the ensemble?
Assume that the models represent the mean and variability correctly, only use trend from observations.
Is the ensemble mean within the error bars of the observations?
Assume that the model mean is correct, and obtain the
variability from the short observational record with a normality assumption.
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Is the ensemble mean within the error bars of the observations?
Compute z-values from the regression estimates and their errors:
z = Aobs − Amod
(∆Aobs)2 + (∆Amod)2/N
(2)
with N the number of ensemble members, the bar denotes the
ensemble average. The standard errors ∆A are computed assuming a normal distribution of the trends A.
The normal approximation has been verified in the 17-member
ECHAM5 ensemble, where the skewness of the 17 trend estimates is less than 0.2 in almost all areas where z > 2.
Serial correlations have been taken into account whenever significant.
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Trends in the Netherlands, observations vs model
0 1 2 3 4 5 6 7 8
-1 -0.5 0 0.5 1 1.5 2 2.5 3 De Bilt obervations
CNT
CMIP3 mean Essence mean
De Bilt (homogenised) A = 2.50 ± 0.39 Central Netherlands Temp. A = 2.33 ± 0.38
Essence A = 1.24 ± 0.09 ⇒ z = 2.8 (p ∼ 0.012)
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Is the ESSENCE mean trend within the error bars of the observations?
DJF MAM
JJA SON
Contours at z = ±2, 3, 4
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Is the ESSENCE mean trend within the error bars of the observations?
DJF MAM
JJA SON
Contours at z = ±2, 3, 4
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Is the observed trend within the PDF of the ensemble?
Compute the fraction q of models that have lower trends the the observed one
q = N + 1/2
1 + Nmod (3)
with N the number of models in the Nmod = 22 model ensemble that have a trend lower than the observed one.
If there are Nrun > 1 runs for one model each run contributes 1/Nrun
to N , so that the models are given equal weight.
Purple (q > 0.975) indicates that the observed trend is higher than all runs of all CMIP3 models simulate, in the red areas
(0.95 < q < 0.975) one run of one model has a higher trend.
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Trends in the Netherlands, observations vs model
0 1 2 3 4 5 6 7 8
-3 -2 -1 0 1 2 3
De Bilt obervations CRUTEM3/HadSST2 CMIP3 ensemble Essence ensemble Essence mean GFDL CM2.1 MIROC 3.2 T106 HadGEM1
CCCMA CGCM 3.1
De Bilt (homogenised) A = 2.50 ± 0.39 Central Netherlands Temp. A = 2.23 ± 0.39
Above all 17 Essence ensemble members (p ∼ 0.028) Above all 5 selected CMIP3 models
Above whole CMIP3 ensemble except one ensemble member (p ∼ 0.036)
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Is the observed trend within the PDF of the CMIP3 ensemble?
a DJF b MAM
c JJA d SON
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Is the observed trend within the PDF of the CMIP3 ensemble?
a DJF b MAM
c JJA d SON
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Is the observed trend within the PDF of a UKMO QUMP ensemble?
a DJF b MAM
c JJA d SON
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Is the observed trend within the PDF of a UKMO QUMP ensemble?
a DJF b MAM
c JJA d SON
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Higher resolution
Observations:
• CRU TS 3
• ENSEMBLES 1.1 Models:
• ENSEMBLES RT2b: RCMs with GCM boundaries
• ENSEMBLES RT3: RCMs with ERA-40 boundaries
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Observed trends do not depend greatly on the resolution
CRUTEM3 1950-2007 CRU TS 3 1950-2006
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Observed trends do not depend greatly on the resolution
CRUTEM3 1950-2007 ENSEMBLES 1950-2006
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Model trends do not depend greatly on the resolution
HadCM3 Q0 HadRM3 Q0
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Model trends do not depend greatly on the resolution
HadCM3 Q16 HadRM3 Q16
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Differences between observed and modelled temperature trends
• Do not depend strongly on resolution
• Highly significant in some seasons and regions
• Not due to weather or decadal climate fluctuations there
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Physical causes of the differences in Europe
• Atmospheric circulation
• Ocean circulation
• Short-wave radiation (aerosols, clouds)
• Soil moisture
• Snow cover
• . . .
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Atmospheric circulation trends
Following van Oldenborgh & van Ulden (2003) and van Ulden & van Oldenborgh (2006) we construct a Very Simple Model (VSM) at each grid point
T(t) = ATglobal (t) + Bu(t) + M T(t − 1) + η(t) (4) with u(t) = (ugeo, vgeo, Vgeo) the geostrophic wind components:
zonal, meridional and vorticity. M is a memory term, and the remainder is classified as noise η(t), time t in months.
Define circulation (in)dependent temperature anomalies:
Tcirc (t) = Bu(t) (5)
Tind (t) = ATglobal (t) + η(t) (6) Geostrophic winds and vorticity are computed from NCEP/NCAR reanalysis SLP.
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Trends in T
circDJF MAM
JJA SON
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Trends in circulation
• Over most of Europe, the trends in the circulation-independent parts are largest in spring and summer; in winter the trends in the circulation dominate.
DJF DJF
NCEP/NCAR SLP trend ECHAM5/MPI-OM SLP trend
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
North Atlantic Ocean circulation bias
u(10m)
750m
z=0 u dz
SODA 1.4.1 ECHAM5/MPI-OM
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Section at 52
◦N
SODA 1.4.1 ECHAM5/MPI-OM
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
SST warming trends
SODA-POP 1958-2004 ESSENCE 1958-2004
ESSENCE 1950-2100
o K
in n
jk li k
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Modelled propagation to land
DJF MAM
JJA SON
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
North Atlantic Ocean circulation bias
• The wrong flow of the North Atlantic ocean currents causes too deep mixed layers in the East Atlantic.
• SST trends are much smaller than observed in this area, also in other models.
• The bias in SST trends extends to the Atlantic coasts of Europe.
Note that the ensemble MOC decorrelates within 10 years, so all model decadal variations are sampled.
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Short-wave radiation in spring and summer
Global short-wave observations in the Netherlands are reasonably reliable from the early 1970s onwards. Corrected for circulation.
-30 -20 -10 0 10 20 30 40 50
1970 1980 1990 2000 2010 2020 2030 Essence ensemble
De Bilt observations
Wageningen observations
JJA
• Aerosol dip around 1985 in observations and model?
• Linear trend 1971-2007 in model and observations?
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Aerosol effects
0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20
1940 1960 1980 2000 2020 2040 2060 2080 2100 2120
Dec-Feb
0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20 0.22 0.24
1940 1960 1980 2000 2020 2040 2060 2080 2100 2120
Mar-May
0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50
1940 1960 1980 2000 2020 2040 2060 2080 2100 2120
Jun-Aug
0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50
1940 1960 1980 2000 2020 2040 2060 2080 2100 2120
Sep-Nov
16.00 18.00 20.00 22.00 24.00 26.00 28.00
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005
Dec-Feb
12.00 14.00 16.00 18.00 20.00 22.00 24.00
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005
Mar-May
10.00 12.00 14.00 16.00 18.00 20.00 22.00
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005
Jun-Aug
15.00 16.00 17.00 18.00 19.00 20.00 21.00 22.00 23.00 24.00 25.00 26.00
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005
Sep-Nov
Essence sulphate column Number of days with vv<5km
(Vertical) (Horizontal)
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Aerosol effects in Holland
• In summer, observed aerosol concentrations in 2000 are roughly equal to those in 1950, with a peak in the early 1980s.
• In the model, summer concentrations are much lower in 2000 than in 1950.
• Hence, the model should overestimate the effect on the trend 1950-2000.
• However, comparing radiation measurements, it underestimates the effect of aerosols.
We estimate that the model underestimates the summer effects of aerosols by a factor 2 to 3. Projected on Tglobal(3) , this translates to about 20% of the discrepancy in trends.
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Snow melting around the Baltic
NOAA snow cover ECHAM5/MPI-OM
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Snow melting around the Baltic
0 0.2 0.4 0.6 0.8 1
-0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5
Mar-May averaged index
Mar-May averaged HadCRUT3_global_temperature
NOAA snowcover 10-30E 55-65N vs HadCRUT3 global temperature 1972:2007
3.1972 1973 1975 1974 1976
19771978
1979 1980
1981
1982 1983 1984
1985
1986 1987 1988
1989 1990 1991 19931992 19951994
1996
1997 1998
19992000 2001
2002 2003
2004 2005 2006
2007
r = -0.358
0 0.2 0.4 0.6 0.8 1
287 287.2 287.4 287.6 287.8 288 288.2 288.4 288.6
Mar-May averaged index
Mar-May averaged Essence Tglobal
Essence (ECHAM5/MPI-OM) snow cover 10-30E 55-65N vs Essence Tglobal 1972:2007 r = -0.042
NOAA snow cover ECHAM5/MPI-OM
The difference is about 95% significant (taking a 3-yr autocorrelation into account).
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Relative precipitation trends 1901-2007 in the CMIP3 ensemble
0 2 4 6 8 10 12 14
-40 -20 0 20 40
winter relative precipitation trends Nertherlands (13 stations) CMIP3 ensemble
0 2 4 6 8 10 12 14
-40 -20 0 20 40
summer relative precipitation trends Nertherlands (13 stations) CMIP3 ensemble
October–March [%/K] April–September [%/K]
Multiple ensemble members of the same model have been weighed by 1/Nens so that each model contributes equally to the histogram.
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Comparison of observed trend with CMIP3
summer
winter
Observed rel. trend Mean modelled trend Quantile of obs (GPCC) [1/K] (CMIP3) [1/K] in model PDF Also noted by Zhang et al, Nature, 2007.
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Comparison of observed trend with CMIP3
October–March April–September
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Comparison with a perturbed physics ensemble
summer
winter
Observed rel trend Mean modelled trend Quantile of obs (GPCC) [1/K] (QUMP) [1/K] in model PDF
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Uninitialised climate simulations: conclusions 1
• Global warming is now strong enough to start verifying climate models on the temperature and precipitation trends.
• Although global and continental-scale warming are simulated well by climate models, there are large biases on regional scales.
• In coastal Europe, the warming is underestimated by a factor 1.5 to 2 so far (up to 3σ): very unlikely only due to chance weather and decadal climate fluctuations.
• In northern Europe, the trend towards more precipitation in winter is underestimated by a factor two or more.
• These biases are not caused by the coarse resolution: RCMs show the same biases.
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Uninitialised climate simulations: conclusions 2
• In winter and early spring the largest changes are in the
circulation, with sea-level pressure in the Mediterranean rising much faster than modelled.
• A wrong circulation in the North Atlantic Ocean in most climate models causes an underestimation of the warming trends along the Atlantic coast.
• In late spring and summer, the observed trend in short-wave radiation in the Netherlands is simulated much weaker by the ECHAM5 model. The decrease in aerosols only explain a small part of the trend 1950-2007, but seems underestimated by the models.
• In spring, snow melt trends are much smaller than observed in the Baltic
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Initialised climate simulations
• Up to now we only considered the skill from boundary
conditions: increasing GHG concentrations, volcanic aerosols, tropospheric aerosols.
• How do we fare when including the initial state: mainly the ocean, maybe cryosphere, land surface.
Here we look at the skill in four ENSEMBLES models (IFS33r1,
HadGEM2, ARPEGE4.6, ECHAM5) that performed 10-yr hindcasts starting Nov 1 1960, 1965, 1970, . . . , 2005.
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Global mean temperature
-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
1880 1900 1920 1940 1960 1980 2000 no volcanoes
with volcanoes GISS Ts obs HadCRUT3 obs Mauna Loa CO2
• Without volcanoes, Tglobal follows the CO2 concentration
• The largest deviations over 1958–2008 are due to volcanoes and internal variability. Part of the latter may be predictable.
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Global mean temperature forecasts
-0.4 -0.2 0 0.2 0.4 0.6 0.8
-0.4 -0.2 0 0.2 0.4 0.6 0.8
Tglobal predicted
Tglobal observed year 1
r=0.97 a=0.65+/-0.05 RMSE=0.041 K
-0.4 -0.2 0 0.2 0.4 0.6 0.8
-0.4 -0.2 0 0.2 0.4 0.6 0.8
Tglobal predicted
Tglobal observed years 2-5
r=0.92 a=0.75+/-0.12 RMSE=0.072 K
-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8
-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8
Tglobal predicted
Tglobal observed Year 6-10
r=0.982 a=0.87+/-0.07
RMSE=0.036 K mean all
IFS33r1 HadGEM2 ARPEGE4.6 ECHAM5
• The ENSEMBLES decadal runs have good skill in Tglobal
• Mainly due to the trend
• The CMIP3 runs without volcanoes have trend 0.80 ± 0.06 against observed Tglobal for single years, 0.94 ± 0.04 for 3-yr means, 0.99 ± 0.09 for 5-yr means. The trends in the decadal runs are 10% lower. Bias in trend / trend in bias.
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Correlation skill SST globally
• The ENSEMBLES ensemble has good skill in SST almost
everywhere (exception: areas of doubtful observations in NCDC ERSST v3b).
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Land 2m temperature globally
• The ENSEMBLES ensemble has good skill in T2m almost
everywhere (exception: areas of doubtful observations in NCEP GHCN/CAMS).
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Global mean temperature forecasts minus trend
-0.4 -0.2 0 0.2 0.4
-0.4 -0.2 0 0.2 0.4
Tglobal predicted
Tglobal observed year 1 without trend
r=0.93 a=0.51+/-0.07 RMSE=0.031 K
-0.4 -0.2 0 0.2 0.4
-0.4 -0.2 0 0.2 0.4
Tglobal predicted
Tglobal observed years 2-5 without trend
r=-0.02 a=-0.03+/-0.44 RMSE=0.057 K
-0.4 -0.2 0 0.2 0.4
-0.4 -0.2 0 0.2 0.4
Tglobal predicted
Tglobal observed Year 6-10 without trend
r=0.612 a=0.82+/-0.43
RMSE=0.036 K mean all
IFS33r1 HadGEM2 ARPEGE4.6 ECHAM5
• Define the trend as projection on the CO2 concentration
• Good skill in year 1 mainly due to persistence and ENSO (forecasts start Nov 1).
• No skill in yr 2–5 (r = −0.01)
• Non-significant skill in yr 6–10 (r = 0.29, p = 0.3)
• (Note: not yet cross-validated)
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
SST globally minus trend
yr 2–5 yr 6–10
• Good skill in North Atlantic yr 2–5, some left in yr 6–10.
• Reasonable skill in decadal ENSO region in yr 6–10, maybe PDO.
Why re-emergence?
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Land 2m temperature minus trend
yr 2–5 yr 6–10
• Much less skill over land: maybe Alaska, Central Asia
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Atlantic Multidecadal Oscillation
-0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5
1880 1900 1920 1940 1960 1980 2000 2020
AMO [C]
Jan-Dec AMO ersst (amo_ersst_ts)
SST EQ–60◦N, 80◦– 0◦W minus SST 60◦S–60◦N (Trenberth and Shea 2006)
• Big shifts around 1995, 1970 and 1930, maybe 2010.
• Almost orthogonal to global warming over 1880–now and
1960–now (also in ensembles ECHAM5/MPI-OM and CCSM 3.0 relative to ensemble mean)
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
AMO forecasts
-0.4 -0.2 0 0.2 0.4 0.6 0.8
1950 1960 1970 1980 1990 2000 2010
AMO
year 1 RMSE=0.081 K a=0.48+/-0.19 r=0.67
-0.4 -0.2 0 0.2 0.4 0.6 0.8
1950 1960 1970 1980 1990 2000 2010
AMO
years 2-5 RMSE=0.050 K a=0.74+/-0.18 r=0.84
-0.4 -0.2 0 0.2 0.4 0.6 0.8
1960 1970 1980 1990 2000 2010
AMO
Year 6-10
r=0.591 IFS33r1
HadGEM2 ARPEGE4.6 ECHAM5 multimodel mean observations
• The ENSEMBLES decadal runs show good skill in forecasting the AMO variations: drop after 1960, rise around 1995.
• In yr 2–5 r = 0.84, in yr 6–10 still r = 0.59.
• They do not reproduce the shift around 1995 except in year 1.
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
AMO teleconnections: T2m (without trend)
AMO teleconnection Multimodel skill yr 2–5
• Teleconnections are somewhat similar to skill ENSEMBLES runs: regions with a positive or negative teleconnection have overlap with the regions with positive skill
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
AMO teleconnections: precipitation
AMO teleconnection Multimodel skill yr 2–5
• Teleconnections are somewhat similar to skill ENSEMBLES runs: regions with a positive or negative teleconnection have overlap with the regions with positive skill
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
AMO teleconnections: precipitation (w/o trend)
AMO teleconnection Multimodel skill yr 2–5
• Teleconnections are somewhat similar to skill ENSEMBLES runs: regions with a positive or negative teleconnection have overlap with the regions with positive skill
n o
in k
li jk
s M e n d a rl e d e N
te o ro lo
g is c h
In s ti tu
u
Conclusions
• In temperature most skill comes from the boundary forcing:
global warming.
• The fluctuations around the trend of the global mean
temperature are not predicted well by the ENSEMBLES decadal runs (IFS, ARPEGE, ECHAM5, HadGEM2).
• The models have good skill in hindcasts of the low-frequency behaviour of SST in the North Atlantic around the trend.
• The skill in temperature and precipitation after taking the trend into account overlaps with AMO teleconnections.
• Hence the ocean initial state contributes significantly to skill in these regions.
