Role of hydrological model uncertainties in climate change impact - - PowerPoint PPT Presentation

▶
role of hydrological model uncertainties in climate
SMART_READER_LITE
LIVE PREVIEW

Role of hydrological model uncertainties in climate change impact - - PowerPoint PPT Presentation

Apr 10, 2023 480 likes •699 views

Role of hydrological model uncertainties in climate change impact studies Satish Bastola, Conor Murphy, and John Sweeney ICARUS, NUIM Ireland HydroPredict 2010: 20 -23 September 2010 Prague, Czech Republic Contents Introduction

slide-1
SLIDE 1

Role of hydrological model uncertainties in climate change impact studies

Satish Bastola, Conor Murphy, and John Sweeney

ICARUS, NUIM Ireland

HydroPredict’ 2010: 20-23 September 2010 Prague, Czech Republic

slide-2
SLIDE 2

Contents

  • Introduction
  • Method

– Account for hydrological model uncertainty – quantify uncertainty in impact studies

  • Results
  • Conclusion
slide-3
SLIDE 3

Uncertainty that cascade through a climate change impact assessment

Emission Scenarios: Economic activity, population growth, Technology Response of climate model to emissions Impact models (simple eqn, spa/tem aggr, calib,plausable

  • Projected changes in future climate are inherently uncertain
  • Considerable work have focused on Emission and GCM

uncertainty but have mostly neglected uncertainties in impact models

slide-4
SLIDE 4

Objective

Examine the role of model uncertainty (parameter and structural uncertainty) in climate change impact studies

  • A. Account for hydrological model uncertainty

(GLUE, BMA)

  • B. Quantify uncertainties that cascade through

the climate change impact assessment

slide-5
SLIDE 5

Schematic: accounting for uncertainty in Hydrological model using GLUE

GLUE method to account for Para & Str.Uncertainty in Hydrological models

Model 1 Model 2 Model k GLUE: Generalized Likelihood Uncertainty Estimation Method (Beven and Binley, 1992) L : Likelihood; θ: Model parameters; TH: threshold of Likelihood GLUE has been extensively used (e.g. Freer et al., 1996; Freer et al., 2004; Montanari, 2005 and more)

) / ( 1 ) | (

2 2

  • bs

i i Y

L     

Time

Streamflow

Time (2000-2100)

Simulators pdf of parameter

Parameter θ Likelihood L Threshold TH

θ L

TH

θ L

TH

) , ( I f Y  

θ

L

θ

L

θ

L

Simulation

slide-6
SLIDE 6

Bayesian Model averaging

M1 Mk Δ

Time

In BMA the predictive probability density function (PDF) of any quantity of interest is a weighted average of PDFs centred on the individual forecasts

p(Δ|M1) p(Δ|Mk)

x x x x (Tutorial on BMA: Hoeting et al., 1999)

  

K k k k K

D M p D M p D M M p

1 1

) | ( ) , | ( ) , ,.., | (

Posterior distribution

  • f BMA prediction

Weight (Wk)

slide-7
SLIDE 7

Bayesian Model Averaging

For BMA Time (2000-2100)

)) | ( .. ) | ( ) | ( log( ) , ,.. | ,.. (

2 2 1 1 1 2 2 1 1 k k n t k k

M P w M P w M P w w w l       

 

Weight and variance parameter of BMA were estimated using DREAM of Vrugt et al (2008). Model 1 Model 2 Model k GLUE

  

K k k k K

W D M p D M M p

1 1

) , | ( ) , ,.., | (

For each conditional PDF gamma distribution was selected

  

 ) ( |

) / ( 1

   

   e

M k

  • k

k k k k k k k k

c M b M      . ; / ; /

2 2 2 2

       

W1,W2..Wk, b, Co

slide-8
SLIDE 8

Study Region & Data

Irish National Meteorological Service

  • Office of Public works, ireland

Six regional climate scenarios from Fealy and Sweeney, 2007

Republic of Ireland A2 B2

HadCM3 CCCMA CSIRO HadCM3 CCCMA CSIRO

CCCma (CGCM2):canadian centre for climate modeling and analysis; CSIRO: Commonwealth Scientific and Industrial Research Organization; HadCM3:Hadley Centre Coupled Model, version 3

slide-9
SLIDE 9

Hydrological model

CQOF,TOF,TIF, TG

TANK (Sugawara, 1995) HyMOD NAM (DHI)

m a x

( 1 )

r z a p r

S E E S  

(Beven,1991)

u z i d

S q v S Dt 

Beven and wood,1993) Beven,1984

TOP Model (Beven et al1995)

The Hymod, NAM, TANK, and TOP models describes the behaviour of each individual component in the hydrological cycle, at catchment level, by using a set of mathematical equations.

slide-10
SLIDE 10

Model uncertainty using GLUE/BMA

Hymod NAM Tank Top Simulated flow:Daily (3yr)/seasonal(1971-1991)

prediction quantile (Cal (1971-1989):Multimodel) BOYNE 50 100 150 200 250 300 350 1/1/81 1/4/81 1/7/81 1/10/81 1/1/82 1/4/82 1/7/82 1/10/82 1/1/83 1/4/83 1/7/83 1/10/83 Time Streamflow (Cumecs)

parameter & model structural uncertainty

prediction quantile (Cal (1971-1989):Hymod) BOYNE 50 100 150 200 250 300 350 1/1/81 1/3/81 1/5/81 1/7/81 1/9/81 1/11/81 1/1/82 1/3/82 1/5/82 1/7/82 1/9/82 1/11/82 1/1/83 1/3/83 1/5/83 1/7/83 1/9/83 1/11/83 Time Streamflow (Cumecs) prediction quantile (Cal (1971-1989):TOP BOYNE 20 40 60 80 100 120 140 160 180 1/1/81 1/3/81 1/5/81 1/7/81 1/9/81 1/11/81 1/1/82 1/3/82 1/5/82 1/7/82 1/9/82 1/11/82 1/1/83 1/3/83 1/5/83 1/7/83 1/9/83 1/11/83 Time Streamflow (Cumecs) prediction quantile (Cal (1971-1989):NAM) BOYNE 50 100 150 200 250 300 350 1/1/81 1/3/81 1/5/81 1/7/81 1/9/81 1/11/81 1/1/82 1/3/82 1/5/82 1/7/82 1/9/82 1/11/82 1/1/83 1/3/83 1/5/83 1/7/83 1/9/83 1/11/83 Time Streamflow (Cumecs) prediction quantile (Cal (1971-1989):Tank) BOYNE 50 100 150 200 250 300 350 1/1/81 1/3/81 1/5/81 1/7/81 1/9/81 1/11/81 1/1/82 1/3/82 1/5/82 1/7/82 1/9/82 1/11/82 1/1/83 1/3/83 1/5/83 1/7/83 1/9/83 1/11/83 Time Streamflow (Cumecs)

Accounts for parameter uncertainty Whymod, b, Co WNam, b, Co WTank, b, Co WTOP, b, Co BMA

HYMOD_MOY 50 100 150 1 3 5 7 9 11 month (climatological axis) Cumecs

Range med Obs

NAM_MOY 50 100 150 1 3 5 7 9 11 month (climatological axis) Cumecs

Range med Obs

Tank_MOY 50 100 150 1 3 5 7 9 11 month (climatological axis) Cumecs

Range med Obs

TOP_MOY 50 100 150 1 3 5 7 9 11 month (climatological axis) Cumecs

Range med Obs

GLUE BMA

slide-11
SLIDE 11

Model calibration/validation

PI (Width of 90 Prediction interval); CE (No of points with in PI/No of points)

Basin (Model) NSE (Median) Calib Valid Calib Valid Calib Valid 1 Moy (HYMOD) 0.77 0.66 0.68 0.56 30.50 33.01 2 Moy (NAM) 0.72 0.63 0.58 0.52 25.69 27.66 3 Moy (TANK) 0.80 0.69 0.80 0.77 40.88 44.55 4 Moy (TOP) 0.80 0.70 0.72 0.70 33.98 37.47 0.81 0.72 0.85 0.80 43.32 46.84 5 Boyne (HYMOD) 0.79 0.76 0.80 0.83 28.17 29.35 6 Boyne(NAM) 0.76 0.74 0.77 0.78 23.82 25.10 7 Boyne (TANK) 0.70 0.73 0.67 0.75 25.60 27.13 8 Boyne (TOP) 0.69 0.68 0.52 0.57 23.26 24.74 0.80 0.78 0.90 0.92 31.78 33.40 9 Suck (HYMOD) 0.78 0.68 0.70 0.68 17.27 18.75 10 Suck (NAM) 0.72 0.63 0.56 0.51 14.68 15.85 11 Suck (TANK) 0.70 0.65 0.61 0.59 17.08 18.45 12 Suck (TOP) 0.68 0.60 0.34 0.31 12.65 14.06 0.79 0.69 0.74 0.70 19.24 20.92 13 Blackwater (HYMOD) 0.64 0.74 0.50 0.58 25.18 25.67 14 Blackwater (NAM) 0.63 0.72 0.31 0.40 15.62 16.13 15 Blackwater (TANK) 0.67 0.75 0.59 0.68 33.35 34.09 16 Blackwater (TOP) 0.64 0.71 0.33 0.31 21.77 22.69 0.66 0.74 0.68 0.76 36.52 37.32 Ensemble Med Ensemble Med Ensemble Med Ensemble Med 1971-1990/1991- 2000 1971-1990/1991- 2000 1971-1990/1991- 2000 1971-1990/1991- 2000

Sn Period (Calib/Valid)

CE PI (m3/s)

slide-12
SLIDE 12

GLUE/BMA

BMA GLUE

slide-13
SLIDE 13

Quantify uncertainties that cascade through the climate change impact assessment

B

slide-14
SLIDE 14

Uncertainty Envelope: Experiment design

CCCma CSIRO HadCM3 A2 B2 A2 B2 A2 B2

HyMod Tank NAM TOP

HyMod Tank NAM TOP HyMod Tank NAM TOP HyMod Tank NAM TOP HyMod Tank NAM TOP HyMod Tank NAM TOP

Hydro

Hydro: Hydrological model uncertainty (parameter & model selection) Scenario: Hydrological + Scenario (A2 & B2) Scenario GCM GCM: Hydrological + Scenario (A2 & B2) Total: Uncertainty envelop (Hydrological + Scenario (A2 & B2)+GCM) Total GCM: Weighted based on Climate prediction index Scenarios: Equally likely Model: Equally Likely Simulators: Weighted based on Likelihood value w1,σ1; w2,σ2; w3,σ3; w4,σ4 (The weight parameters are revised based on GCM weight)

BMA

slide-15
SLIDE 15

Hydrological model uncertainty

The average width of the PI from parameterization of CRR models is nearly 50% 10 20 30 40 50 60 70 80 90 100

CCCMA (A2) CCCMA (B2) CISRO (A2) CISRO (B2) HADCM3 (A2) HADCM3 (B2) CCCMA (A2) CCCMA (B2) CISRO (A2) CISRO (B2) HADCM3 (A2) HADCM3 (B2) CCCMA (A2) CCCMA (B2) CISRO (A2) CISRO (B2) HADCM3 (A2) HADCM3 (B2) CCCMA (A2) CCCMA (B2) CISRO (A2) CISRO (B2) HADCM3 (A2) HADCM3 (B2)

Widh of prediction interval (% of average flow) Climate scenarios

a)2050-2059

Moy Boyne Blackwater suck

HYMOD NAM TANK TOP Uncertainty due to parameterization and Model selection

20 40 60 80 100 120 2020s 2050s 2070s 2020s 2050s 2070s Parameterization Model selection Prediction interval (% of average flow)

Moy Boyne Blackwater Suck

and nearly increased to 70% when Different CRR models are included

slide-16
SLIDE 16

Apportion of uncertainty

25 50 75 100 125 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 Average prediction width (% of average obseved flow)

HYDRO

a)2020s

25 50 75 100 125 150 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 Average prediction width (% of average obseved flow)

HYDRO SCENE

a)2020s 25 50 75 100 125 150 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 Average prediction width (% of average obseved flow)

HYDRO SCENE GCM(A2) GCM (B2)

a)2020s 25 50 75 100 125 150 175 200 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 Average prediction width (% of average obseved flow)

HYDRO SCENE GCM(A2) GCM (B2) Total (GLUE) Total (BMA)

a)2020s

The average width of the PI is 70% (of average streamflow) when uncertainty in hydrological response to single GCM was quantified. This increased to 100% when two SRES scenarios were employed. Further increases to 120% when three GCM with single scenarios was used, and further to 140% when two SRES scenarios were used.

25 50 75 100 125 150 175 200 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 HADCM3_A2 HADCM3_B2 CCCMA_A2 CCCMA_B2 CSIRO_A2 CSIRO_B2 Average prediction width (% of average obseved flow) HYDRO SCENE GCM(A2) GCM (B2) Total (GLUE) Total (BMA)

a)2020s

Blackwater Boyne Moy Suck

slide-17
SLIDE 17

BMA/GLUE

20 40 60 80 100 120 140 160 180

HYDRO SCENE GCM (A2) GCM (B2) TOTAL

Experiment design

Average prediction width (% of average observed flow) GLUE BMA

a)2020s

50 100 150 200

HYDRO SCENE GCM (A2) GCM (B2) TOTAL

Experiment design

c) 2070s

50 100 150 200

HYDRO SCENE GCM (A2) GCM (B2) TOTAL

Experiment design

b) 2050s

For BMA widths of Prediction interval are higher than GLUE by a factor of 1.4, 1.2, 1.2, 1.2, 1.1 for HYDRO, SCENE, GCMA, GCMB, Total respectively.

slide-18
SLIDE 18

Comparison of BMA and GLUE estimated Prediction interval

d)BOYNE:2020s

50 100 150

J F M A M J J A S O N D Streamflow (Cumecs)

e)BOYNE: 2050s

50 100 150

J F M A M J J A S O N D

f) BOYNE: 2070s

50 100 150

J F M A M J J A S O N D

a)Blackwater: 2020s

50 100 150 200 250

J F M A M J J A S O N D Streamflow (Cumecs)

90% Prediction range (GLUE) Median (GLUE) Upper 95% (BMA) Lower 5% (BMA) Median (BMA)

b)Blackwater: 2050s

50 100 150 200 250 J F M A M J J A S O N D

c)Blackwater: 2070s

50 100 150 200 250

J F M A M J J A S O N D

Blackwater Boyne

slide-19
SLIDE 19

Conclusion

  • This study is an attempt to quantify the uncertainty in the

projection of future water resources by incorporating four plausible yet conceptually diverse models forced with six regional climate change scenarios, using BMA and GLUE.

  • Both GLUE and BMA approaches used here differ

fundamentally in their underlying philosophy and representation of error.

  • The role of hydrological model uncertainty is considerable

and warrants inclusion in impacts assessment, particularly where robust adaptation decisions are required.

  • When A2 and B2 SRES scenarios are considered, the GCM

uncertainty was observed to be higher than emission uncertainty.

  • Results are indicative as the full range of emission

scenarios and GCM sensitivities were not sampled here.

slide-20
SLIDE 20

Thank you very much for your attention