Sippel4IMSC biac correction
A novel bias correction methodology for climate impact simulations Sebastian Sippel1 1
Max Planck Institute for Biogeochemistry, Jena, Germany
06.06.2015
... and many more people: Miguel Mahecha, Jakob Zscheischler, Friederike Otto, Matthias Forkel, Fabian Gans, Myles Allen, Martin Heimann, Markus Reichstein, Sonia Seneviratne
Sippel et al. (MPI-BGC)
06.06.2015
1 / 19
Outline
1. Bias correction for climate impact simulations
2. Application to sub-seasonal time scales
3. Extending ‘updating technique’ to multi-model ensembles
Sippel et al. (MPI-BGC)
06.06.2015
2 / 19
1. Bias correction for climate impact simulations
Sippel et al. (MPI-BGC)
06.06.2015
3 / 19
Mean Warming = + 0.5 [SD] Prob (X > x) = 0.01
2
3
4
5
No bias in scale Relative Bias (Scale) = 75 % Relative Bias (Scale) = 150 %
1
Return level [arbitrary units]
a)
6
Relevance of biases
10
100
200
500
15
150
10
100 75
Relative bias in scale [%]
20
200
50
0
Probability ratio
50
Return time [arbitrary units]
5
b)
20
0.0
0.5
1.0
1.5
2.0
Mean Warming [SD]
Sippel et al. (MPI-BGC)
06.06.2015
4 / 19
Drawbacks: Statistical bias correction
I
Severe manipulation of the original data
I
Changes to correlation structure are induced
I
Physical inconsistency
I
...
Sippel et al. (MPI-BGC)
06.06.2015
5 / 19
A novel ensemble bias correction scheme
Idea: Select those ensemble members that are plausible given the observed distribution of any meteorological variable.
Sippel et al. (MPI-BGC)
06.06.2015
6 / 19
A novel ensemble bias correction scheme b
Central Europe, areamean 100
● ● ● ● ● ●
0.8
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
0.6
0.4
0.2
0.0 14
16
18
Model Ensemble, Percentiles
Cumulative Density Function
1.0
15.5
80
60
40
20
ERA−Interim Gaussian Kernel, Obs Gaussian Kernel, Mod
20
22
24
Temperature, JJA mean [°C]
0
Observations, Tair [°C] 16.5 17 17.5 18
19
Hermite Splines ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
0
20
40
60
80
20.5 20 19.5 19 18.5 18 17.5 17 16.5 16 15.5
Model Ensemble, Tair [°C]
a
100
Observations, Percentiles
Sippel et al., 2016, ESD
Advantages of resampling bias correction: I No data manipulation I Multivariate correlation structure remains intact Disadvantage: I Size of the ensemble decreases I Resampling means that any ensemble member might be chosen more than once Sippel et al. (MPI-BGC)
06.06.2015
7 / 19
Evaluation of bias correction: C. Europe b
26
Precipitation sum, JJA [°C]
Mean temperature, JJA [°C]
a
24 22 20 18 16 14
500 400 300 200 100 0
ORIG
PROBCOR
ISIMIP
OBS
ORIG
PROBCOR
ISIMIP
OBS
ORIG
PROBCOR
ISIMIP
OBS
d Mean incoming LW radiation [W m−2]
Mean incoming SW radiation [W m−2]
c 350
300
250
200
150 ORIG
PROBCOR
ISIMIP
OBS
380 370 360 350 340 330 320
Sippel et al., 2016, ESD Sippel et al. (MPI-BGC)
06.06.2015
8 / 19
Climatic extremes
●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●● ●● ● ● ● ●●●● ● ● ●● ● ● ●●● ● ● ● ● ●●● ● ● ● ● ●●● ●●● ● ● ● ●● ● ● ●● ● ● ●●●● ● ●● ●● ● ●● ●●●● ●● ● ●●● ● ●● ●● ● ●● ●● ● ●●● ● ●● ●●● ●●●● ●● ●● ●●● ●● ●●●●● ●● ●● ●● ●●● ● ●● ●●●● ● ●● ●●●●● ●● ●● ●● ● ●● ●● ●● ● ● ●●● ●●● ● ●● ●● ● ● ● ●● ●●●● ● ●● ●● ●● ● ●● ●● ● ●● ● ● ● ● ● ●● ● ● ●● ● ●●●● ● ●● ● ● ● ●● ●● ●● ●●● ● ● ● ● ● ●●● ● ● ● ● ● ●●● ● ●
b
35
Central Europe, area mean, JJA Tair, monthly
25 ●● ●● ● ●
●●●● ●
●● ● ●●● ●●●● ● ● ● ●● ● ● ● ●● ● ● ● ●●● ●● ●● ●● ●● ●●●● ● ● ●● ● ●
20
18
Bias correction, 7 obs. T datasets ● ●
16
30
Temperature [°C]
Temperature [°C]
a
●●● ●● ● ●
● ●● ●●● ● ● ● ●
14
● ● ● ●●● ● ●
●● ●●● ● ●
●●●● ● ●
12 10
15
8 10
20
50
100
200
500 1000
10
20
50
100
● ● ●
0
●
10
20
50 100
400
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●● ● ● ● ●●●● ● ●●●●●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●●●● ● ● ● ● ●●●●●●● ● ●● ● ●●●● ●● ●●● ● ●●●●● ●●●●●●● ●● ● ● ● ●●●●●● ●●●●●●●● ●●●●●● ●● ● ●● ● ● ●●●●●●●● ●● ● ●●●●● ● ● ● ● ● ● ● ●● ●● ●● ●● ●
●
HadRM3P−ORIG PROBCOR Conventional bias cor. Observations
500
2000
300
●● ● ●
200
●● ● ● ●
500 1000
Central Europe, area mean, JJA cumulative precipitation
● ● ●●
● ● ●● ●
100
●● ● ● ●
200
●
200
● ●● ●
●● ● ● ●
0
● ● ●●
Precipitation, JJA [mm]
500 400 300
Precipitation, JJA [mm]
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●●●●●● ● ●● ● ●● ● ●● ●● ●● ●●● ●● ● ●●● ●●●● ●●● ●● ●●● ● ● ● ● ●●● ● ●● ●●● ●● ●● ●● ● ●●●●●●●●●● ●●● ●●● ● ●● ●●● ●● ●●● ● ●●● ● ● ● ●●●●●●●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●●●● ●●●●
100
Return time [years] b
500
Return time [years] a
HadRM3P−ORIG PROBCOR ISIMIP Observations
10
20
Return time [years]
50 100
500
2000
Return time [years]
Sippel et al., 2016, ESD
Sippel et al. (MPI-BGC)
06.06.2015
9 / 19
2. Application to sub-seasonal time scales
Sippel et al. (MPI-BGC)
06.06.2015
10 / 19
40 35
JJA Tair (3d, seas. max) [°C]
35 30 25
25
JJA Tair (3d, seas. max) [°C]
20
30
b)
a)
45
Applicability on daily time scales?
2
35
18
20
22
5
10
20
50
100 200
500
28
Return time [yr]
26
24
JJA daily max. WBT (3d, seas. max) [°C]
26
d)
24
30
20
16
JJA daily max. WBT (3d, seas. max) [°C]
25
21 day seas. max temperature [°C]
22
20
c)
15
20
25
30
21 day seas. max temperature [°C]
35
2
5
10
20
50
100 200
500
Return time [yr]
Sippel et al, in preparation
Sippel et al. (MPI-BGC)
06.06.2015
11 / 19
3. Extending ‘updating technique’ to multi-model ensembles
Sippel et al. (MPI-BGC)
06.06.2015
12 / 19
Land coupling in ‘bias-corrected’ ensemble a
HadRM3P − original
b
HadRM3P − bias. cor.
65°N
65°N
60°N
60°N
55°N
55°N
50°N
50°N
45°N
45°N
10°W 0°
c
10°E
20°E
10°W
30°E
0°
EOBS − Upscaled LE
d
10°E
20°E
30°E
ERA−Interim
65°N
65°N
60°N
60°N
55°N
55°N
50°N
50°N
45°N
45°N
10°W 0°
10°E −1.0
20°E
10°W
30°E
−0.5
0°
0.0
0.5
10°E
20°E
30°E
1.0
Kendall rank correlation
Sippel et al. (MPI-BGC)
06.06.2015
13 / 19
Introduction: Vegetation-land-atmosphere coupling
Zscheischler et al., 2015 (GRL) I Advantages of VAC: I I
Differentiates between categories Coincidence analysis of events allows comparison of different datasets
I Disadvantages I
Statistical approach, i.e. no physics-based information about feedback strengths, etc. Sippel et al. (MPI-BGC)
06.06.2015
13 / 19
Vegetation-land-atmosphere coupling
Zscheischler et al., 2015 (GRL) Figure 1. (a) Feedbacks between T , ET/FPAR, and SM in energy- and SM-limited regimes. + and − denote precipitation. VACa and VACb are associated with strong anomalies in T and ET/FPAR with the same sign. opposite signs. The dashed line between T and P/clouds corresponds to the negative atmospheric covari moisture can also affect precipitation (either positively or negatively) in some regions (see text). (b) FPAR samples in Europe. Points are colored according to the classes of VACFPAR . (c) As in Figure 1b but for Aust Sippel et al. (MPI-BGC)
06.06.2015
14 / 19
1.0
1. How is land-coupling represented in climate models?
●
CEU
0.4
VACc
0.6
0.8
●
Diagnostic LSM Reanalyses Median Diagnostic (LFE) Median LSM (LFE) Median Reanalyses (LFE) LandFluxEVAL−Median
●
●
0.2
●
●
● ●
● ● ● ●
●
●
●
●
JJA−OBS
JJA−CMIP5
MAM−OBS
MAM−CMIP5
All−OBS
Randomness All−CMIP5
0.0
● ● ●
1989-2005
Sippel et al. (MPI-BGC)
06.06.2015
15 / 19
1. How is land-coupling represented in climate models?
● ●●
● ●
●
CEU−JJA
OBS
●
NorESM1−ME NorESM1−M MRI−ESM1 MRI−CGCM3 MPI−ESM−MR MPI−ESM−LR MIROC5 MIROC−ESM IPSL−CM5B−LR IPSL−CM5A−MR IPSL−CM5A−LR inmcm4 HadGEM2−ES HadGEM2−CC HadGEM2−AO GISS−E2−R GISS−E2−H GFDL−ESM2M GFDL−ESM2G GFDL−CM3 FIO−ESM FGOALS−g2 EC−EARTH CSIRO−Mk3 CNRM−CM5 CMCC−CMS CMCC−CM CMCC−CESM CESM1−CAM5 CESM1−BGC CCSM4 CanESM2 bcc−csm1−1−m bcc−csm1−1 ACCESS1−3 ACCESS1−0
● ● ●● ● ●
●
●●
● ●
●
● ●
● ●
● ●
●
●
●
● ●
●
●
●
●● ● ●
● ● ●
●
● ● ●
●
●
● ● ●
●
● ● ● ● ●
0.0 Sippel et al. (MPI-BGC)
●
0.2
0.4 VACc
●
0.6
0.8 06.06.2015
16 / 19
1. How is land-coupling represented in climate models?
● ● ● ● ●
●
● ●
● ● ●
● ●
● ● ● ● ● ●
● ● ●
●
●
●
● ●
● ●
● ● ● ●
● ● ● ● ● ● ●
●
●
●
● ●
● ●
● ● ● ● ● ● ●
● ●
● ●
● ● ● ● ●
● ● ● ●
●
● ● ● ● ●
●
● ● ●
● ● ● ● ●
● ●
● ● ● ●
● ● ● ● ● ● ● ● ●
● ● ● ●
● ● ● ● ●
● ● ● ●
●
● ●
● ● ● ● ●
●
● ● ● ●
● ● ● ● ●
●
● ●
●
● ● ● ● ● ●
●
● ●
● ● ● ●
●
● ●
●
●
●
●
● ●
● ● ● ● ●
● ● ● ● ●
● ● ● ● ● ●
● ● ● ● ●
● ●
● ●
●
●
● ●
● ●
● ● ● ●
●
●
● ● ● ●
● ● ● ● ● ●
● ● ●
● ● ●
● ●
● ●
● ●
● ● ●
● ● ● ● ● ●
● ● ● ● ● ●
●
● ● ●
● ●
●
● ● ● ● ● ● ● ●
● ● ● ● ●
● ● ●
● ● ●
●
● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ●
● ●
● ●
● ● ● ● ●
● ● ● ● ●
● ● ● ● ●
● ● ● ● ● ● ●
●
● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ●
● ● ● ● ● ● ●
● ● ● ● ●
● ● ● ● ● ● ● ● ● ●
● ● ● ● ●
● ● ● ●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ●
●
● ● ● ●
● ● ● ● ●
● ● ● ●
● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ●
● ● ●
● ● ● ● ● ● ● ● ● ●
● ● ● ●
● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
●
● ● ● ●
● ● ● ● ● ● ● ● ● ●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
●
● ● ● ● ● ● ●
● ● ●
● ● ● ●
● ● ● ● ● ●
● ● ● ●
● ●
● ● ●
● ● ● ● ●
● ● ● ● ● ● ● ●
● ● ● ●
● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ●
● ●
● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ●
● ● ●
● ● ● ● ● ● ● ● ● ● ● ●
● ●
● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ●
● ● ● ●
● ● ● ● ● ● ●
● ● ● ● ●
●
● ● ●
● ● ● ● ● ●
● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ●
● ● ● ●
● ● ●
● ● ● ● ● ● ● ● ●
●
●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ●
● ● ● ● ● ●
● ● ●
● ● ●
●
● ●
● ●
●
● ● ● ●
● ● ●
● ● ● ● ● ● ● ●
●
● ● ● ● ●
●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ●
●
● ● ● ● ● ● ● ● ● ●
● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ●
● ● ●
● ●
● ● ●
●
●
● ● ●
● ● ● ● ● ●
●
● ● ● ● ● ●
● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ●
●
● ● ● ● ● ● ● ●
●
● ●
● ●
●
●
●
●
● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ●
● ●
● ●
● ●
●
● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ●
●
● ● ●
● ●
● ● ● ●
● ●
● ● ●
● ●
● ● ●
● ●
●
● ● ●
● ● ● ● ● ● ● ●
● ●
● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ●
●
● ● ● ● ● ● ● ● ●
●
● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ●
● ● ●
● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ●
●
● ● ● ● ● ●
● ● ●
● ●
● ●
● ● ● ● ● ● ● ● ● ● ● ●
●
● ● ● ● ● ● ●
●
● ● ● ●
● ●
● ●
● ● ●
● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ●
● ● ● ● ●
●
● ● ● ● ● ● ●
● ● ● ● ● ● ●
●
● ●
●
● ● ● ● ● ● ● ● ●
●
−0.5 −0.3 −0.1 0.1 0.3 0.5 (Median) VACc_CMIP5 − VACc_LFE
1989-2005 Sippel et al. (MPI-BGC)
06.06.2015
17 / 19
2. Coupling as a model constraint? monthly temperatures: CEU
Tair (CRU) − ET (LandFluxEVAL) observations ● CMIP5
4
●
● ● ● ●
● ●● ● ●
●
3
● ● ● ● ● ● ●
2
●
1
● ●
●
●● ●●
●
● ●● ●●
●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ●●● ●● ●● ●●● ● ● ●●● ● ● ● ●●●
● ● ●
● ●● ● ● ●●
● ●● ●● ● ● ●
●
● ●
●
● ●
●
R = 0.75
0
Variance of T anom. (JJA, °C)
5
●
0.0
0.1
Sippel et al. (MPI-BGC)
0.2
0.3
0.4 VACc
0.5
0.6
0.7 06.06.2015
18 / 19
2. Coupling as a model constraint?
txx mean reduction, 2020: −1.7°C txx 95th perc. reduction, 2020: −3.6°C txx var. reduction, 2020: −48.8% txx mean reduction, 2050: −2°C txx 95th perc. reduction, 2050: −3.2°C txx var. reduction, 2050: −44.7%
CEU
35
● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●●● ● ● ● ●● ●● ● ● ● ● ● ● ● ●● ● ●
30
● ●
25
txx, JJA [°C]
40
45
monthly temperatures:
1960
1980
Sippel et al. (MPI-BGC)
2000
CMIP5, mean bias: 1.8°C CMIP5, var. bias: 8.8°C CMIP5−constr, mean bias: 0.4°C CMIP5−constr, var. bias: 3.1°C
2020 Year
2040
2060
2080 06.06.2015
19 / 19
Conclusions
I
A resampling-based bias correction retains physical consistency and the multivariate correlation structure of regional climate model ensembles
I
Correcting biases by resampling improves the simulation of climate extremes and impacts
I
Could be extended to physics-based constraints, such as land-coupling
Sippel et al. (MPI-BGC)
06.06.2015
20 / 19
Thanks for the attention!
Sippel et al. (MPI-BGC)
06.06.2015
20 / 19
Introduction: Vegetation-land-atmosphere coupling
Zscheischler et al., 2015 (GRL) I Advantages of VAC: I I
Differentiates between categories Coincidence analysis of events allows comparison of different datasets
I Disadvantages I
Statistical approach, i.e. no physics-based information about feedback strengths, etc. Sippel et al. (MPI-BGC)
06.06.2015
20 / 19
Description of bias correction procedure: 1. Estimate the observed probability distribution of any observed meteorological variable (‘constraint’) using e.g. a kernel density estimate. 2. Estimate the probability distribution of the meteorological ‘constraint’ in the large ensemble (or rank the ensemble based on the constraint). 3. Derive a transfer function that maps the observed kernel with the simulated ensemble ranks based on the meteorological ‘constraint’ (e.g. splines). 4. Derive a new (‘bias-corrected’) ensemble by resampling from the observed kernel and retaining the corresponding ensemble members. a
b
Central Europe, areamean ● ● ● ● ●
0.8
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
0.6
0.4
0.2 ●
0.0 14
16
18
80
ERA−Interim Gaussian Kernel, Obs Gaussian Kernel, Mod
20
22
Temperature, JJA mean [°C]
Sippel et al. (MPI-BGC)
60
40
20 ●
Obs. Kernel Percentiles 10 30 60 90
0
100
100
●
Ensemble Percentile
Cumulative Density Function
1.0
24
0
● ● ● ● ● ● ● ● ●● ● ●● ● ●●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ●●● ●
●
●
Hermite Splines
15
16
17
18
19
20
21
Observations [°C]
06.06.2015
19 / 19
Max Planck Institute for Biogeochemistry, Jena, Germany
06.06.2015
... and many more people: Miguel Mahecha, Jakob Zscheischler, Friederike Otto, Matthias Forkel, Fabian Gans, Myles Allen, Martin Heimann, Markus Reichstein, Sonia Seneviratne
Sippel et al. (MPI-BGC)
06.06.2015
1 / 19
Outline
1. Bias correction for climate impact simulations
2. Application to sub-seasonal time scales
3. Extending ‘updating technique’ to multi-model ensembles
Sippel et al. (MPI-BGC)
06.06.2015
2 / 19
1. Bias correction for climate impact simulations
Sippel et al. (MPI-BGC)
06.06.2015
3 / 19
Mean Warming = + 0.5 [SD] Prob (X > x) = 0.01
2
3
4
5
No bias in scale Relative Bias (Scale) = 75 % Relative Bias (Scale) = 150 %
1
Return level [arbitrary units]
a)
6
Relevance of biases
10
100
200
500
15
150
10
100 75
Relative bias in scale [%]
20
200
50
0
Probability ratio
50
Return time [arbitrary units]
5
b)
20
0.0
0.5
1.0
1.5
2.0
Mean Warming [SD]
Sippel et al. (MPI-BGC)
06.06.2015
4 / 19
Drawbacks: Statistical bias correction
I
Severe manipulation of the original data
I
Changes to correlation structure are induced
I
Physical inconsistency
I
...
Sippel et al. (MPI-BGC)
06.06.2015
5 / 19
A novel ensemble bias correction scheme
Idea: Select those ensemble members that are plausible given the observed distribution of any meteorological variable.
Sippel et al. (MPI-BGC)
06.06.2015
6 / 19
A novel ensemble bias correction scheme b
Central Europe, areamean 100
● ● ● ● ● ●
0.8
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
0.6
0.4
0.2
0.0 14
16
18
Model Ensemble, Percentiles
Cumulative Density Function
1.0
15.5
80
60
40
20
ERA−Interim Gaussian Kernel, Obs Gaussian Kernel, Mod
20
22
24
Temperature, JJA mean [°C]
0
Observations, Tair [°C] 16.5 17 17.5 18
19
Hermite Splines ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
0
20
40
60
80
20.5 20 19.5 19 18.5 18 17.5 17 16.5 16 15.5
Model Ensemble, Tair [°C]
a
100
Observations, Percentiles
Sippel et al., 2016, ESD
Advantages of resampling bias correction: I No data manipulation I Multivariate correlation structure remains intact Disadvantage: I Size of the ensemble decreases I Resampling means that any ensemble member might be chosen more than once Sippel et al. (MPI-BGC)
06.06.2015
7 / 19
Evaluation of bias correction: C. Europe b
26
Precipitation sum, JJA [°C]
Mean temperature, JJA [°C]
a
24 22 20 18 16 14
500 400 300 200 100 0
ORIG
PROBCOR
ISIMIP
OBS
ORIG
PROBCOR
ISIMIP
OBS
ORIG
PROBCOR
ISIMIP
OBS
d Mean incoming LW radiation [W m−2]
Mean incoming SW radiation [W m−2]
c 350
300
250
200
150 ORIG
PROBCOR
ISIMIP
OBS
380 370 360 350 340 330 320
Sippel et al., 2016, ESD Sippel et al. (MPI-BGC)
06.06.2015
8 / 19
Climatic extremes
●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●● ●● ● ● ● ●●●● ● ● ●● ● ● ●●● ● ● ● ● ●●● ● ● ● ● ●●● ●●● ● ● ● ●● ● ● ●● ● ● ●●●● ● ●● ●● ● ●● ●●●● ●● ● ●●● ● ●● ●● ● ●● ●● ● ●●● ● ●● ●●● ●●●● ●● ●● ●●● ●● ●●●●● ●● ●● ●● ●●● ● ●● ●●●● ● ●● ●●●●● ●● ●● ●● ● ●● ●● ●● ● ● ●●● ●●● ● ●● ●● ● ● ● ●● ●●●● ● ●● ●● ●● ● ●● ●● ● ●● ● ● ● ● ● ●● ● ● ●● ● ●●●● ● ●● ● ● ● ●● ●● ●● ●●● ● ● ● ● ● ●●● ● ● ● ● ● ●●● ● ●
b
35
Central Europe, area mean, JJA Tair, monthly
25 ●● ●● ● ●
●●●● ●
●● ● ●●● ●●●● ● ● ● ●● ● ● ● ●● ● ● ● ●●● ●● ●● ●● ●● ●●●● ● ● ●● ● ●
20
18
Bias correction, 7 obs. T datasets ● ●
16
30
Temperature [°C]
Temperature [°C]
a
●●● ●● ● ●
● ●● ●●● ● ● ● ●
14
● ● ● ●●● ● ●
●● ●●● ● ●
●●●● ● ●
12 10
15
8 10
20
50
100
200
500 1000
10
20
50
100
● ● ●
0
●
10
20
50 100
400
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●● ● ● ● ●●●● ● ●●●●●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●●●● ● ● ● ● ●●●●●●● ● ●● ● ●●●● ●● ●●● ● ●●●●● ●●●●●●● ●● ● ● ● ●●●●●● ●●●●●●●● ●●●●●● ●● ● ●● ● ● ●●●●●●●● ●● ● ●●●●● ● ● ● ● ● ● ● ●● ●● ●● ●● ●
●
HadRM3P−ORIG PROBCOR Conventional bias cor. Observations
500
2000
300
●● ● ●
200
●● ● ● ●
500 1000
Central Europe, area mean, JJA cumulative precipitation
● ● ●●
● ● ●● ●
100
●● ● ● ●
200
●
200
● ●● ●
●● ● ● ●
0
● ● ●●
Precipitation, JJA [mm]
500 400 300
Precipitation, JJA [mm]
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●●●●●● ● ●● ● ●● ● ●● ●● ●● ●●● ●● ● ●●● ●●●● ●●● ●● ●●● ● ● ● ● ●●● ● ●● ●●● ●● ●● ●● ● ●●●●●●●●●● ●●● ●●● ● ●● ●●● ●● ●●● ● ●●● ● ● ● ●●●●●●●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●●●● ●●●●
100
Return time [years] b
500
Return time [years] a
HadRM3P−ORIG PROBCOR ISIMIP Observations
10
20
Return time [years]
50 100
500
2000
Return time [years]
Sippel et al., 2016, ESD
Sippel et al. (MPI-BGC)
06.06.2015
9 / 19
2. Application to sub-seasonal time scales
Sippel et al. (MPI-BGC)
06.06.2015
10 / 19
40 35
JJA Tair (3d, seas. max) [°C]
35 30 25
25
JJA Tair (3d, seas. max) [°C]
20
30
b)
a)
45
Applicability on daily time scales?
2
35
18
20
22
5
10
20
50
100 200
500
28
Return time [yr]
26
24
JJA daily max. WBT (3d, seas. max) [°C]
26
d)
24
30
20
16
JJA daily max. WBT (3d, seas. max) [°C]
25
21 day seas. max temperature [°C]
22
20
c)
15
20
25
30
21 day seas. max temperature [°C]
35
2
5
10
20
50
100 200
500
Return time [yr]
Sippel et al, in preparation
Sippel et al. (MPI-BGC)
06.06.2015
11 / 19
3. Extending ‘updating technique’ to multi-model ensembles
Sippel et al. (MPI-BGC)
06.06.2015
12 / 19
Land coupling in ‘bias-corrected’ ensemble a
HadRM3P − original
b
HadRM3P − bias. cor.
65°N
65°N
60°N
60°N
55°N
55°N
50°N
50°N
45°N
45°N
10°W 0°
c
10°E
20°E
10°W
30°E
0°
EOBS − Upscaled LE
d
10°E
20°E
30°E
ERA−Interim
65°N
65°N
60°N
60°N
55°N
55°N
50°N
50°N
45°N
45°N
10°W 0°
10°E −1.0
20°E
10°W
30°E
−0.5
0°
0.0
0.5
10°E
20°E
30°E
1.0
Kendall rank correlation
Sippel et al. (MPI-BGC)
06.06.2015
13 / 19
Introduction: Vegetation-land-atmosphere coupling
Zscheischler et al., 2015 (GRL) I Advantages of VAC: I I
Differentiates between categories Coincidence analysis of events allows comparison of different datasets
I Disadvantages I
Statistical approach, i.e. no physics-based information about feedback strengths, etc. Sippel et al. (MPI-BGC)
06.06.2015
13 / 19
Vegetation-land-atmosphere coupling
Zscheischler et al., 2015 (GRL) Figure 1. (a) Feedbacks between T , ET/FPAR, and SM in energy- and SM-limited regimes. + and − denote precipitation. VACa and VACb are associated with strong anomalies in T and ET/FPAR with the same sign. opposite signs. The dashed line between T and P/clouds corresponds to the negative atmospheric covari moisture can also affect precipitation (either positively or negatively) in some regions (see text). (b) FPAR samples in Europe. Points are colored according to the classes of VACFPAR . (c) As in Figure 1b but for Aust Sippel et al. (MPI-BGC)
06.06.2015
14 / 19
1.0
1. How is land-coupling represented in climate models?
●
CEU
0.4
VACc
0.6
0.8
●
Diagnostic LSM Reanalyses Median Diagnostic (LFE) Median LSM (LFE) Median Reanalyses (LFE) LandFluxEVAL−Median
●
●
0.2
●
●
● ●
● ● ● ●
●
●
●
●
JJA−OBS
JJA−CMIP5
MAM−OBS
MAM−CMIP5
All−OBS
Randomness All−CMIP5
0.0
● ● ●
1989-2005
Sippel et al. (MPI-BGC)
06.06.2015
15 / 19
1. How is land-coupling represented in climate models?
● ●●
● ●
●
CEU−JJA
OBS
●
NorESM1−ME NorESM1−M MRI−ESM1 MRI−CGCM3 MPI−ESM−MR MPI−ESM−LR MIROC5 MIROC−ESM IPSL−CM5B−LR IPSL−CM5A−MR IPSL−CM5A−LR inmcm4 HadGEM2−ES HadGEM2−CC HadGEM2−AO GISS−E2−R GISS−E2−H GFDL−ESM2M GFDL−ESM2G GFDL−CM3 FIO−ESM FGOALS−g2 EC−EARTH CSIRO−Mk3 CNRM−CM5 CMCC−CMS CMCC−CM CMCC−CESM CESM1−CAM5 CESM1−BGC CCSM4 CanESM2 bcc−csm1−1−m bcc−csm1−1 ACCESS1−3 ACCESS1−0
● ● ●● ● ●
●
●●
● ●
●
● ●
● ●
● ●
●
●
●
● ●
●
●
●
●● ● ●
● ● ●
●
● ● ●
●
●
● ● ●
●
● ● ● ● ●
0.0 Sippel et al. (MPI-BGC)
●
0.2
0.4 VACc
●
0.6
0.8 06.06.2015
16 / 19
1. How is land-coupling represented in climate models?
● ● ● ● ●
●
● ●
● ● ●
● ●
● ● ● ● ● ●
● ● ●
●
●
●
● ●
● ●
● ● ● ●
● ● ● ● ● ● ●
●
●
●
● ●
● ●
● ● ● ● ● ● ●
● ●
● ●
● ● ● ● ●
● ● ● ●
●
● ● ● ● ●
●
● ● ●
● ● ● ● ●
● ●
● ● ● ●
● ● ● ● ● ● ● ● ●
● ● ● ●
● ● ● ● ●
● ● ● ●
●
● ●
● ● ● ● ●
●
● ● ● ●
● ● ● ● ●
●
● ●
●
● ● ● ● ● ●
●
● ●
● ● ● ●
●
● ●
●
●
●
●
● ●
● ● ● ● ●
● ● ● ● ●
● ● ● ● ● ●
● ● ● ● ●
● ●
● ●
●
●
● ●
● ●
● ● ● ●
●
●
● ● ● ●
● ● ● ● ● ●
● ● ●
● ● ●
● ●
● ●
● ●
● ● ●
● ● ● ● ● ●
● ● ● ● ● ●
●
● ● ●
● ●
●
● ● ● ● ● ● ● ●
● ● ● ● ●
● ● ●
● ● ●
●
● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ●
● ●
● ●
● ● ● ● ●
● ● ● ● ●
● ● ● ● ●
● ● ● ● ● ● ●
●
● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ●
● ● ● ● ● ● ●
● ● ● ● ●
● ● ● ● ● ● ● ● ● ●
● ● ● ● ●
● ● ● ●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ●
●
● ● ● ●
● ● ● ● ●
● ● ● ●
● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ●
● ● ●
● ● ● ● ● ● ● ● ● ●
● ● ● ●
● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
●
● ● ● ●
● ● ● ● ● ● ● ● ● ●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
●
● ● ● ● ● ● ●
● ● ●
● ● ● ●
● ● ● ● ● ●
● ● ● ●
● ●
● ● ●
● ● ● ● ●
● ● ● ● ● ● ● ●
● ● ● ●
● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ●
● ●
● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ●
● ● ●
● ● ● ● ● ● ● ● ● ● ● ●
● ●
● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ●
● ● ● ●
● ● ● ● ● ● ●
● ● ● ● ●
●
● ● ●
● ● ● ● ● ●
● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ●
● ● ● ●
● ● ●
● ● ● ● ● ● ● ● ●
●
●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ●
● ● ● ● ● ●
● ● ●
● ● ●
●
● ●
● ●
●
● ● ● ●
● ● ●
● ● ● ● ● ● ● ●
●
● ● ● ● ●
●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ●
●
● ● ● ● ● ● ● ● ● ●
● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ●
● ● ●
● ●
● ● ●
●
●
● ● ●
● ● ● ● ● ●
●
● ● ● ● ● ●
● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ●
●
● ● ● ● ● ● ● ●
●
● ●
● ●
●
●
●
●
● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ●
● ●
● ●
● ●
●
● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ●
●
● ● ●
● ●
● ● ● ●
● ●
● ● ●
● ●
● ● ●
● ●
●
● ● ●
● ● ● ● ● ● ● ●
● ●
● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ●
●
● ● ● ● ● ● ● ● ●
●
● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ●
● ● ●
● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ●
●
● ● ● ● ● ●
● ● ●
● ●
● ●
● ● ● ● ● ● ● ● ● ● ● ●
●
● ● ● ● ● ● ●
●
● ● ● ●
● ●
● ●
● ● ●
● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ●
● ● ● ● ●
●
● ● ● ● ● ● ●
● ● ● ● ● ● ●
●
● ●
●
● ● ● ● ● ● ● ● ●
●
−0.5 −0.3 −0.1 0.1 0.3 0.5 (Median) VACc_CMIP5 − VACc_LFE
1989-2005 Sippel et al. (MPI-BGC)
06.06.2015
17 / 19
2. Coupling as a model constraint? monthly temperatures: CEU
Tair (CRU) − ET (LandFluxEVAL) observations ● CMIP5
4
●
● ● ● ●
● ●● ● ●
●
3
● ● ● ● ● ● ●
2
●
1
● ●
●
●● ●●
●
● ●● ●●
●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ●●● ●● ●● ●●● ● ● ●●● ● ● ● ●●●
● ● ●
● ●● ● ● ●●
● ●● ●● ● ● ●
●
● ●
●
● ●
●
R = 0.75
0
Variance of T anom. (JJA, °C)
5
●
0.0
0.1
Sippel et al. (MPI-BGC)
0.2
0.3
0.4 VACc
0.5
0.6
0.7 06.06.2015
18 / 19
2. Coupling as a model constraint?
txx mean reduction, 2020: −1.7°C txx 95th perc. reduction, 2020: −3.6°C txx var. reduction, 2020: −48.8% txx mean reduction, 2050: −2°C txx 95th perc. reduction, 2050: −3.2°C txx var. reduction, 2050: −44.7%
CEU
35
● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●●● ● ● ● ●● ●● ● ● ● ● ● ● ● ●● ● ●
30
● ●
25
txx, JJA [°C]
40
45
monthly temperatures:
1960
1980
Sippel et al. (MPI-BGC)
2000
CMIP5, mean bias: 1.8°C CMIP5, var. bias: 8.8°C CMIP5−constr, mean bias: 0.4°C CMIP5−constr, var. bias: 3.1°C
2020 Year
2040
2060
2080 06.06.2015
19 / 19
Conclusions
I
A resampling-based bias correction retains physical consistency and the multivariate correlation structure of regional climate model ensembles
I
Correcting biases by resampling improves the simulation of climate extremes and impacts
I
Could be extended to physics-based constraints, such as land-coupling
Sippel et al. (MPI-BGC)
06.06.2015
20 / 19
Thanks for the attention!
Sippel et al. (MPI-BGC)
06.06.2015
20 / 19
Introduction: Vegetation-land-atmosphere coupling
Zscheischler et al., 2015 (GRL) I Advantages of VAC: I I
Differentiates between categories Coincidence analysis of events allows comparison of different datasets
I Disadvantages I
Statistical approach, i.e. no physics-based information about feedback strengths, etc. Sippel et al. (MPI-BGC)
06.06.2015
20 / 19
Description of bias correction procedure: 1. Estimate the observed probability distribution of any observed meteorological variable (‘constraint’) using e.g. a kernel density estimate. 2. Estimate the probability distribution of the meteorological ‘constraint’ in the large ensemble (or rank the ensemble based on the constraint). 3. Derive a transfer function that maps the observed kernel with the simulated ensemble ranks based on the meteorological ‘constraint’ (e.g. splines). 4. Derive a new (‘bias-corrected’) ensemble by resampling from the observed kernel and retaining the corresponding ensemble members. a
b
Central Europe, areamean ● ● ● ● ●
0.8
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
0.6
0.4
0.2 ●
0.0 14
16
18
80
ERA−Interim Gaussian Kernel, Obs Gaussian Kernel, Mod
20
22
Temperature, JJA mean [°C]
Sippel et al. (MPI-BGC)
60
40
20 ●
Obs. Kernel Percentiles 10 30 60 90
0
100
100
●
Ensemble Percentile
Cumulative Density Function
1.0
24
0
● ● ● ● ● ● ● ● ●● ● ●● ● ●●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ●●● ●
●
●
Hermite Splines
15
16
17
18
19
20
21
Observations [°C]
06.06.2015
19 / 19
