A framework for visualizing the convergence performance of … · 2020. 5. 2. · A framework for...

A framework for visualizing the convergence

performance of global optimization algorithms for

hydrological models

Tian Lan, Kairong Lin, Chong-Yu Xu, and XiaohongChen

https://meetingorganizer.copernicus.org/EGU2020/EGU2020-1792.html

Hanzhong basin Mumahe basin Xunhe basin

Yellow River

Wei River

Yangtze RiverChengdu Plain

Hangjiang River

3461

0

2531

437

2961

226

Miles Miles Miles

Extraction of dynamic

catchment characteristics

Calibration of dynamic

parameters

Multi-metric assessment

of dynamic parameters

Indices are specified with dynamic

catchment characteristics.

Multiple clustering operations based

on individual indices-systems.

Calibration period Validation period Individual dynamic

parameters for

calibration

Whole dynamic

parameter set

for calibration

Linear and nonlinear

correlation between

parametersDifferent sub-periods for objective function

• Parameter

Sub-periods

Calibration period

Clustering operation I

Clustering operation II

Pre-processed using

MIC and PCA

Climatic and land-surface indices

VS

Q5 Q20 Q70 Q95

FDC

• RMSE_Q5

• RMSE_Q20

• RMSE_Q70

• RMSE_Qmid

• RMSE_Q95

• NSE

• LNSE

Framework Discussion

Kq

B

alpha

Ks

Huz

a

b

(I) (II) (III)

Unimodal distribution Bimodal distribution Flat distributionMultimodal distribution

(IV)

Pa

ram

ete

r valu

es

High probability

Low probability

cf(x)

0

1

Parameter

space

Objective function values

Local optimum

(I) (II)

(III) (IV)

Global optimum

Evolutional direction

d r = 0.8

MIC = 0.5

r = -0.1

MIC = 0.1

r = -0.8

MIC = 0.5

r = 0.0

MIC = 0.6

r = 0.0

MIC = 0.8

r = 0.0

MIC = 0.0

e

Model performance with time-invariant parameters in calibration period

Model performance with time-invariant parameters in validation period

Model performance with dynamic parameters in calibration period

Model performance with dynamic parameters in validation period

0 0.5 1

1-NSE

1-LNSE

RMSE_Q5

RMSE_Q20

RMSE_mid

RMSE_Q70

RMSE_Q95

1-NSE

1-LNSE

RMSE_Q5

RMSE_Q20

RMSE_mid

RMSE_Q70

RMSE_Q95

0 0.5 1

1-NSE

1-LNSE

RMSE_Q5

RMSE_Q20

RMSE_mid

RMSE_Q70

RMSE_Q95

1-NSE

1-LNSE

RMSE_Q5

RMSE_Q20

RMSE_mid

RMSE_Q70

RMSE_Q95

0 0.5 1

1-NSE

1-LNSE

RMSE_Q5

RMSE_Q20

RMSE_mid

RMSE_Q70

RMSE_Q95

1-NSE

1-LNSE

RMSE_Q5

RMSE_Q20

RMSE_mid

RMSE_Q70

RMSE_Q95


0

0.2

0.4

0.6

0.8

1


NS

E

Dry period Rainfall period I Rainfall period II Rainfall period III


Ks

Kq

alpha

B

Huz

Min Max Min Max Min Max


b

c

d

Hanzhong basin

Mumahe basin

Xunhe basin

1 2 3 4 5 6 7 8 9 10 11 12 13 1914 15 16 17 18 20 21 22 23 24Half month

Dry period Rainfall period I Rainfall period II Rainfall period IIIa

Huz

B

alpha

Kq

Ks


Dry period

Rainfall period I

Rainfall period II

Rainfall period III

f (x) Huz B alpha Kq Ks


0

1

a

b

Hanzhong

basin

Mumahe

basin

Xunhe

basin

Huz B alpha Kq Ks

Parameter valuesOb

jective

fu

nctio

n v

alu

es

A framework for visualizing the convergence performance of … · 2020. 5. 2. · A framework for...

Documents

Transcript of A framework for visualizing the convergence performance of … · 2020. 5. 2. · A framework for...