C. Onyutha , P. Nyeko-Ogiramoi & P. Willems July 28 th , 2014

C. Onyutha, P. Nyeko-Ogiramoi & P. Willems

July 28th, 2014

Uncertainties in Impact Assessment of Climate Change on Rainfall Extremes in the Lake Victoria Basin

2

Why Lake Victoria Basin?WorldLVB has 2nd largest fresh water lake

R. NileKey source

Extreme rainfalls episodes of floods+ loss of lives+ damage to prop.

Population - so high - increased poverty

Need for proper mgt of water resource

3rd International Conference on Earth Science and Climate Change

3

Data

1140 - 1451 1452 - 1763 1764 - 2075 2076 - 2386 2387 - 2698 2699 - 3010 3011 - 3321 3322 - 3633 3634 - 3945 No Data

200 0 200 Kilometers

Rainfall Stations Lake Victoria LVB

DEM [m]

N

Rainfall 1961-2000

GCMs14 – CMIP3 7 – CMIP5

Control/Hist. sim 20 – CMIP3 17 – CMIP5 Fut. Projections 53 – CMIP3 35 – CMIP5

Horizon 2049-2065 [2050s] 2081-2100 [2090s]

Scenarios B1, A1B, A2 – CMIP3 rcp4.5, 6.0, 8.5 – CMIP5

ScenariosB1-550ppm of CO2

A1B-720ppm of CO2

A2-850ppm of CO2


4

Sampling extremes

Inter-GCM differences

Difference between statistical downscaling methods

Sources of uncertainty

Those due to:


5

Data limitation: sampling extremes

Risk based applications: - detailed representation - rainfall

Planning, Design & Operation: accurate descriptive study - extremes &

recurrence rates - long-term series

Extreme value analysis: Data scarce area - short-term series –extrapolation

Predictive uncertainty: - large - important to quantify for decision making

Uncertainty – sampling of extremes


6

Extract independent extremes – AMS/PDS

Quantifying uncertainty How ?

0

10

20

30

40

50

60

70

80

90

100

01/61 06/66 12/71 06/77 11/82 05/88

Ra

infa

ll in

ten

sity [m

m/d

ay]

Time [days]

"Rainfall series"POT valuesPOTs

EVD fitting

Resampling

POTsStation 4


7

Parameter estimation – GPD(loc, scale, shape) - fit distribution - Q-Q plot


POTs

EVD fitting

Resampling

EVD fitting

t = 75

0

20

40

60

80

100

120

0.1 1.0 10.0 100.0

Rai

nfal

l int

ensi

ty [m

m/d

ay]

Return Period [years]

a)t = 75

xt =39.4beta=16.22MSE=0.67

0

5

10

15

20

25

30

35

40

0

5

10

15

20

25

0 20 40 60 80 100

MSE

[mm

/day

]2

Slop

e in

Q-Q

plo

t [m

m/d

ay]

Number of observations above threshold

b)

Method – Weighted Linear Regression*

Station 5

*Onyutha, C. & Willems, P. (2014). Uncertainty in calibrating generalised Pareto distribution to rainfall extremes in Lake

Victoria basin. Hydrol. Res. Doi:10.2166/nh.2014.052


8

Uncertainty on extremes– resampling - Jackknife – at 5% lev. Sig


POTs

EVD fitting

ResamplingResampling

a) Delete ith event from POTs of size w

b) Determine Ɵ(β,γ,xt) of EVD using (w -1) POTs

c) Steps (a) and (b) repeated w times

d) Rank w sets of Ɵ(β,γ,xt) in descending order

e) Upper and lower limits are determined from [0.025× w]th and [0.975× w]th Ɵ(β,γ,xt)

f) 95% CI determined using (e) above


9



POTs

EVD fitting


Station 5


10



POTs

EVD fitting


Station 5

Lower lim

Upper lim


11



POTs

EVD fitting


Station 5

Lower lim

Upper lim


12



POTs

EVD fitting


10 years < 40 data record length(1961-2000)– what if, say 100 years is used?

Station 5

Lower lim

Upper lim10-year quantiles


13

Climate change

&Extremes


14

Climate change: Managers confronted - impact on hydro-clim. Extremes Impact analysis : - not uncertainty-free - large differences exist

Inter-Global climate model differences

Source: IPCC (2013) report

IPCC model global warming projections


15

Inter-compare GCMs: metrics - Bias, SEE, β-related


A higher value of β means higher variations in extreme rainfall intensity

1 10 1000

20

40

60

80

100

120

140

160 a) Station 3- CMIP3AOM_R1BCM2.0_R1CGT47_R1CGT47_R2CGT47_R3CGT63_R1CM2.0_R1CM2.1_R1CM3.0_R1ECH4_R1ECH5_R1ECH5_R4ECHO-G_R1ECHO-G_R2ECHO-G_R3MI3.2H_R1MI3.2M_R1MI3.2M_R2MK3.0_R1Return period [year]

Rai

nfal

l int

ensi

ty [

mm

/day

]

1 10 1000

10

20

30

40

50

60 b) Station 3 - CMIP5BNU_R1MK3.6_R1MK3.6_R2MK3.6_R3MK3.6_R4MK3.6_R5MK3.6_R6MK3.6_R7MK3.6_R8MK3.6_R9MK3.6_R10FG2_R1GFCM_R1GFCM_R3GESM_R1MLR_R1

Return period [year]

Rai

nfal

l int

ensi

ty [

mm

/day

]

Empirical POTs


16

Bias versus SEE: - quantifying differences btn control runs & empirical series


0

20

40

60

0 40 80 120 160

SEE

[mm

/day

]

Absolute Bias [%]

Station 5

0

4

8

12

16

20

0 50 100 150

SEE

[m

m/d

ay]

Absolte Bias [%]

CMIP3CMIP5

Station 5


17

Control versus scenario: β – EVD scale parameter


1

5

25

125

1 5 25 125

βfo

r sce

nario

of t

he 2

050s

β for the control simulations

A2

A1B

B1

1

5

25

125

1 5 25 125

βfo

r sc

enar

io o

f th

e 20

90s

β for the control simulations

A2

A1B

B1

Station 5 – CMIP3


18

Combining empirical, control and scenario:


0.2 2.5 25.00.2

2.5

25.0

A2

A1B

B1

β for empirical over β for control

β fo

r sc

enar

io 2

050s

ove

r β

for

cont

rol

0.2 1.0 5.0 25.00.2

2.5

25.0

A2

A1B

B1

β for empirical over β for control β

for

scen

ario

209

0s o

ver

β fo

r co

ntro

l

Station 5 – CMIP3


19

Control versus scenario: change in β


0.2

1.0

5.0

25.0

0.2 1.0 5.0 25.0C

hang

e [-

] in

βfo

r th

e 20

90s

Change [-] in β for the 2050s

A2

A1B

B1

1

5

25

125

1 5 25 125

βfo

r th

e 20

90s

β for the 2050s

A2

A1B

B1

Station 5 – CMIP3


20

Seive-out the GCMs:

Inconsistency of the GCMs What should I do?

Decision: Downscale – consistent GCMs

Poor performance for:

CMIP5 by GFDL-ESM-2G_R1 at station 9

CMIP3

by at station(s)

CSIRO-MK3.0,R1 4 and 6

GISS-AOM,R1 5 and 9

GFDL-CM2.0,R1 5


21

Difference in statistical downscaling what methods?

Delta: changes in mean of rainfall

PFut,d = future daily rainfall PObs,d = the observed daily seriesPSce,m = mean of GCM scenario series for month mPCon,m = mean of GCM control series for month m


22


Scenario series - quantiles sx1 ≥ sx2 ≥ … ≥ sxi…..≥ sxB; Control series quantiles cx1 ≥ cx2 ≥ … ≥ cxi…..≥ cxB; Empirical quantiles ex1 ≥ ex2 ≥ … ≥ exi…..≥ exB; Quantile pert. factors Qp1 ≥ Qp2 ≥ … ≥ Qpi…..≥ QpB, relative Qpi = sxi/cxi

Quantile based:

Threshold<1 mm/day…AbsQp

Thresh. >1 mm/day…RelQp



(simQP)


23


Scenario series - quantiles sx1 ≥ sx2 ≥ … ≥ sxi…..≥ sxB; Control series quantiles cx1 ≥ cx2 ≥ … ≥ cxi…..≥ cxB; Empirical quantiles ex1 ≥ ex2 ≥ … ≥ exi…..≥ exB; Quantile pert. factors Qp1 ≥ Qp2 ≥ … ≥ Qpi…..≥ QpB, relative Qpi = sxi/cxi

Quantile based:

Threshold<1 mm/day…AbsQp

Thresh. >1 mm/day…RelQp



(simQP)

Wet spell-based: similar to simQP but addition/removal of wet spell


24

Changes in rainfall extremes

-20

-10

0

10

20

30

40

1 2 3 4 5 6 7 8 9

Cha

nge

[%]

Station [-]

a) 2050s-A2DeltasimQPwetQP

-20

-10

0

10

20

30

40

1 2 3 4 5 6 7 8 9C

hang

e [%

]Station [-]

b) 2090s-A2DeltasimQPwetQP

-10

0

10

20

30

1 2 3 4 5 6 7 8 9

Cha

nge

[%]

Station [-]

c) 2050s-A1B DeltasimQPwetQP

-20

-10

0

10

20

30

40

1 2 3 4 5 6 7 8 9

Cha

nge

[%]

Station [-]

d) 2090s-A1B DeltasimQPwetQP

-20

-10

0

10

20

30

1 2 3 4 5 6 7 8 9

Cha

nge

[%]

Station [-]

e) 2050s-B1 DeltasimQPwetQP

-30

-20

-10

0

10

20

30

1 2 3 4 5 6 7 8 9

Cha

nge

[%]

Station [-]

f) 2090s-B1 DeltasimQPwetQP

Changes in rainfall extremes: Change [%] = (xe - xg)/xg*100

where xe = 10-year empirical quantile

xg = 10-year GCM-based quantile

76 m

m/d

ay

70 96 93 9181 90 92 80


Range of change in 10-year rainfall quantile – simQP using CMIP3

Changes in rainfall extremes

-20

-10

0

10

20

30

40

1 2 3 4 5 6 7 8 9

Cha

nge

[%]

Station [-]

2050s-B1

Ensemble mean Maximum Minimum

-20

-10

0

10

20

30

40

50

1 2 3 4 5 6 7 8 9

Cha

nge

[%]

Station [-]

2090s-A2

Ensemble mean Maximum Minimum

100%

76 m

m/d

ay81 70 96 90 93 92 91 80

25


Combining diff. sources of uncertainties

100%

Station 7:

Sampling uncertainty:

upper limit: 118 mm/day

Estimated : 92 mm/day

Lower limit: 84 mm/day

CMIP3 -2090s – A2

Mean Min Max

Delta -9 -24 2

simQP 16 -6 40

wetQP 12 1 3726

Inter-GCM differences – projection of changes[%]:

CMIP5 -2090s – rcp8.5

Mean Min Max

Delta -4 -16 3

simQP 12 -4 29

wetQP 10 -3 22


27

Conclusions

100%

Need to quantify uncertainty in sampling of extremes– data limitation

Inter-GCM differences be used to select those to be downscaled

Choice of a downscaling method – objective driven


[email protected]

….. statistical testing of errors show their presence but not their absence

THANKS FOR LISTENING

C. Onyutha , P. Nyeko-Ogiramoi & P. Willems July 28 th , 2014

Documents

climate change extremes

earth science

rd international conference

rainfall extremes

sigquantifying uncertaintyhow

sampling extremes risk

lower limupper lim

lake victoria basin